Skip to main content

CBE NEXUS · UNDERGRADUATE PHYSICS · UNIT 17

Mathematical Physics

Step-by-step calculations · derivations · transforms · differential equations · mathematical modelling

University Physics / Undergraduate Physics / Unit 17

Learning outcomes

By the end of this unit, students should manipulate complex quantities; test analyticity; solve matrix and eigenvalue problems; evaluate vector derivatives and integrals; construct Fourier series and transforms; solve initial-value problems with Laplace transforms; solve ordinary and partial differential equations; use special functions and Green functions; apply variational principles; and assess numerical accuracy.

Calculation method: For every problem, write the governing equation, state assumptions, substitute numerical or symbolic values, retain units, show intermediate algebra, and check the final result against the original equation and physical dimensions.

Contents

  1. Complex numbers and polar representation
  2. Complex differentiation and analytic functions
  3. Taylor and Laurent series; singularities
  4. Matrices, linear systems and determinants
  5. Eigenvalues, eigenvectors and diagonalisation
  6. Vector algebra and coordinate systems
  7. Gradient, divergence and curl
  8. Line, surface and volume integrals
  9. Laplacian and potential theory
  10. Fourier series and orthogonality
  11. Fourier transform and spectral analysis
  12. Laplace transforms
  13. Ordinary differential equations and physical models
  14. Partial differential equations and separation
  15. Heat equation and diffusion
  16. Wave equation and normal modes
  17. Special functions: Bessel, Legendre and Hermite
  18. Green functions and forced systems
  19. Calculus of variations and Euler–Lagrange equation
  20. Numerical methods and error control
  21. 30 worked calculations
  22. University examination questions
Figure 1. Complex plane and polar representation
z = x + iyx = r cos θy = r sin θθReIm

1. Complex numbers and polar representation

A complex number z=x+iy consists of real part x and imaginary part y, where i²=−1. Its conjugate is z*=x−iy, modulus is |z|=√(x²+y²), and argument is θ=atan2(y,x). The polar representation z=r(cosθ+i sinθ)=re^{iθ} is particularly useful for oscillations, waves and linear systems. Euler's identity e^{iθ}=cosθ+i sinθ follows from the power series of the exponential, sine and cosine. Multiplication adds arguments and multiplies moduli; division subtracts arguments and divides moduli. Complex notation is a compact mathematical representation: measurable physical quantities remain real, usually obtained by taking the real part of a complex solution.

For z₁=2+3i and z₂=1−2i, z₁z₂=(2+3i)(1−2i)=2−4i+3i−6i²=8−i. The modulus of z₁ is √13 and its argument is tan⁻¹(3/2). When using inverse trigonometric functions, identify the quadrant; a plain arctangent can return the wrong angle. De Moivre's theorem (cosθ+i sinθ)^n=cos(nθ)+i sin(nθ) follows by repeated multiplication and is useful for roots and powers.

In AC circuit analysis, a sinusoidal voltage v(t)=V₀cos(ωt+φ) is represented by the phasor V=V₀e^{iφ}, with the time factor e^{iωt} understood. Differentiation in time corresponds to multiplication by iω, and integration corresponds to division by iω for nonzero frequency. This turns differential equations into algebraic equations. Keep a consistent peak or RMS convention throughout a calculation; mixing them introduces a factor √2 error.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

2. Complex differentiation and analytic functions

A complex function f(z) is differentiable at z₀ only if the difference quotient has the same limit for every direction of approach in the complex plane. Writing f=u(x,y)+iv(x,y), differentiability under suitable smoothness conditions requires the Cauchy–Riemann equations u_x=v_y and u_y=−v_x. These conditions are much stronger than ordinary real differentiability. Analytic functions possess derivatives in a neighbourhood and have powerful integral and series properties. Examples include e^z, sin z and rational functions away from their poles.

For f(z)=z², with z=x+iy, f=(x²−y²)+i(2xy). Thus u_x=2x, v_y=2x, u_y=−2y and v_x=2y, so both Cauchy–Riemann equations hold everywhere. The derivative is f'(z)=2z. For f(z)=z*, u=x and v=−y; u_x=1 but v_y=−1, so the function is not complex differentiable. This illustrates why complex differentiability is restrictive.

Analyticity lets one evaluate difficult real integrals using contour methods and residues. It also underpins two-dimensional potential theory: the real and imaginary parts of an analytic function are harmonic, satisfying Laplace's equation u_xx+u_yy=0. In electrostatics and ideal fluid flow, this links complex potentials to field lines and equipotential curves. Boundary conditions and physical singularities determine which analytic function is relevant.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

3. Taylor and Laurent series; singularities

An analytic function near z₀ can be expanded as a Taylor series f(z)=Σ a_n(z−z₀)^n, where a_n=f⁽ⁿ⁾(z₀)/n!. If the function has an isolated singularity, a Laurent series may contain negative powers: f(z)=Σ a_n(z−z₀)^n. A removable singularity has no physically meaningful divergence after redefinition; a pole has a finite principal part; an essential singularity has infinitely many negative-power terms. The residue is the coefficient of (z−z₀)⁻¹.

For f(z)=1/(z−2), the Laurent series about z=0 is not a Taylor series valid near the origin because the nearest singularity is at z=2. For |z|>2, rewrite 1/(z−2)=z⁻¹[1−2/z]⁻¹=Σ_{n=0}∞2ⁿ/zⁿ⁺¹. For |z|<2, write −(1/2)[1−z/2]⁻¹=−Σ zⁿ/2ⁿ⁺¹. The region of convergence is essential information, not an optional detail.

Series expansions approximate functions locally and support perturbation theory. The residue of a simple pole at z=a is lim_{z→a}(z−a)f(z). For a pole of order m, use a derivative formula. Always identify singularities inside the contour and confirm that no pole lies on the integration path unless a principal-value prescription is explicitly used.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

4. Matrices, linear systems and determinants

A matrix represents a linear transformation or a system of coupled equations. For Ax=b, a unique solution exists when det(A)≠0. For a 2×2 matrix A=[[a,b],[c,d]], det(A)=ad−bc and A⁻¹=(1/det A)[[d,−b],[−c,a]]. Gaussian elimination is usually more stable and efficient than computing an inverse explicitly. In physics, matrices describe coupled oscillators, coordinate transformations, rotations, quantum observables and discretised differential operators.

Solve 2x+y=5 and x−y=1. The coefficient matrix is [[2,1],[1,−1]], with determinant −3. Adding the equations gives 3x=6, so x=2 and y=1. Check both original equations. A numerical solution should always be checked by evaluating the residual r=b−Ax; a small residual is necessary but may not guarantee small solution error for an ill-conditioned matrix.

The condition number measures sensitivity of the solution to perturbations. Nearly dependent rows make a system ill-conditioned. Scaling variables, using pivoting, and selecting a better physical parameterisation can improve numerical reliability. In experimental inverse problems, a poorly conditioned matrix often signals that the measurements do not independently constrain the desired parameters.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

5. Eigenvalues, eigenvectors and diagonalisation

An eigenvector v of A satisfies Av=λv, where λ is its eigenvalue. Nonzero solutions exist when det(A−λI)=0, the characteristic equation. If a matrix has a complete basis of independent eigenvectors, it can be diagonalised A=PDP⁻¹. Powers and matrix exponentials then become simpler: e^{At}=Pe^{Dt}P⁻¹. Symmetric real matrices have real eigenvalues and orthogonal eigenvectors, a property central to normal modes and quantum mechanics.

For A=[[2,1],[1,2]], det(A−λI)=(2−λ)²−1=0, giving λ=3 or 1. For λ=3, (A−3I)v=0 gives v proportional to (1,1). For λ=1, v is proportional to (1,−1). Normalised eigenvectors are (1,1)/√2 and (1,−1)/√2. The two normal modes therefore correspond to equal-phase and opposite-phase motion.

Diagonalisation decouples linear coupled equations when the eigenvector basis is complete. The eigenvalues of a dynamical matrix determine squared natural frequencies in stable mechanical systems. Negative squared frequencies indicate instability in the assumed equilibrium. Degeneracy and defective matrices require additional care; do not assume every matrix has a full eigenbasis.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

6. Vector algebra and coordinate systems

Vectors have magnitude and direction. The dot product a·b=ab cosθ is a scalar and measures projection; the cross product a×b has magnitude ab sinθ and direction given by the right-hand rule. Cartesian coordinates are convenient for translational symmetry, cylindrical coordinates for axial symmetry and spherical coordinates for central symmetry. In curvilinear coordinates, basis vectors and scale factors vary with position, so derivatives require geometric terms.

For a=(1,2,2) and b=(2,0,1), a·b=1×2+2×0+2×1=4. The cross product is (2,3,−4), whose magnitude is √29. The angle satisfies cosθ=4/(3√5). Check the cross-product direction using the determinant formula or right-hand rule.

Choose coordinates to match the geometry before differentiating. A radial field around a point source is simpler in spherical coordinates; a long straight wire suggests cylindrical coordinates. Incorrectly using Cartesian derivative formulas with curvilinear components can omit scale factors and produce incorrect physical dimensions.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

7. Gradient, divergence and curl

For scalar field φ, gradient ∇φ points in the direction of steepest increase and has magnitude equal to the maximum directional derivative. For vector field A, divergence ∇·A measures local source strength and curl ∇×A measures local circulation. In Cartesian coordinates, ∇φ=(φ_x,φ_y,φ_z), ∇·A=A_x,x+A_y,y+A_z,z, and curl is the determinant expression involving unit vectors and partial derivatives. These operators have coordinate-dependent expressions in cylindrical and spherical systems.

Let φ=x²y+3z. Then ∇φ=(2xy,x²,3). At (1,2,0), the gradient is (4,1,3). For A=(x²,xy,yz), divergence is 2x+x+z=3x+z. At (1,2,3), it equals 6. State the point after evaluating a field; a gradient field is not itself a scalar potential.

The electrostatic field is E=−∇V, Gauss's law in differential form is ∇·E=ρ/ε₀, and Faraday's law relates curl E to changing magnetic field. These equations show that vector calculus is not abstract decoration: it expresses local physical laws. The divergence theorem and Stokes' theorem convert local differential statements into integral conservation laws.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

8. Line, surface and volume integrals

A line integral accumulates a field along a path; a surface integral measures flux through a surface; a volume integral accumulates a density through a region. For a vector field F, work is ∫F·dr and flux is ∬F·dS. The orientation of a path or surface matters for vector integrals. Conservative fields have path-independent line integrals and zero curl in simply connected regions.

For constant force F=(3,4) N and displacement from (0,0) to (2,1) m, work is F·Δr=3×2+4×1=10 J. For a radial field F=kr r̂, integrate along a radial path from r=a to b: W=∫_a^b kr dr=k(b²−a²)/2.

The divergence theorem states ∬_S F·dS=∭_V∇·F dV for a closed outward-oriented surface. Stokes' theorem states ∮_C F·dr=∬_S(∇×F)·dS. Use these theorems only when the surface, boundary and regularity assumptions are satisfied.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

9. Laplacian and potential theory

The Laplacian of a scalar is ∇²φ=∇·∇φ. In Cartesian coordinates it is φ_xx+φ_yy+φ_zz. Laplace's equation ∇²φ=0 describes source-free regions in electrostatics, steady heat conduction and incompressible irrotational flow. Poisson's equation ∇²φ=f describes a field with sources. Boundary conditions determine a unique physical solution under common well-posed settings.

For φ=x²−y²+z², ∇²φ=2−2+2=2, so it does not satisfy Laplace's equation. For φ=xy+yz+zx, all pure second derivatives vanish, so ∇²φ=0. A harmonic function obeys the mean-value property and cannot have a strict interior maximum or minimum unless constant under the usual conditions.

Separation of variables often reduces Laplace's equation to ordinary differential equations. Geometry dictates the coordinate system and eigenfunctions. Rectangular regions lead to sine and cosine modes; circular and spherical regions lead to Bessel and Legendre functions. Match boundary values carefully and check the resulting solution by substitution.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

10. Fourier series and orthogonality

A periodic function with period 2L can be represented, under suitable conditions, by a Fourier series f(x)=a₀/2+Σ[a_n cos(nπx/L)+b_n sin(nπx/L)]. Orthogonality allows each coefficient to be calculated independently by integration over one period. Even functions contain cosine terms only; odd functions contain sine terms only. At a jump discontinuity, the series converges to the midpoint of the left and right limits, with oscillatory overshoot near the jump described by the Gibbs phenomenon.

For f(x)=x on −π<x<π, odd symmetry gives a_n=0 and a₀=0. The sine coefficients are b_n=(1/π)∫_{−π}^{π}x sin(nx)dx=2(−1)^{n+1}/n. Thus x=2Σ_{n=1}∞(−1)^{n+1}sin(nx)/n for −π<x<π. This series is a standard example of representing a ramp-like periodic waveform.

Fourier methods solve driven oscillations, heat flow and wave propagation by decomposing complicated signals into simple modes. The coefficient convention depends on the chosen period and normalisation. State the interval and period before writing a Fourier series, otherwise factors of π or 2 can be misplaced.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

11. Fourier transform and spectral analysis

The Fourier transform converts a function of position or time into its spatial-frequency or temporal-frequency representation. One common convention is F(k)=∫_{−∞}^{∞}f(x)e^{−ikx}dx and f(x)=(1/2π)∫F(k)e^{ikx}dk. Other conventions distribute factors of 2π differently. Differentiation becomes multiplication by ik, convolution becomes multiplication in transform space, and translation produces a phase factor.

For f(x)=e^{−a|x|}, a>0, even symmetry gives F(k)=2∫₀∞e^{−ax}cos(kx)dx=2a/(a²+k²). This Lorentzian spectrum shows that a sharp spatial or temporal decay has a broad frequency distribution. Verify transform conventions before comparing formulae across texts.

In experiments, finite sampling and finite observation windows modify the ideal transform. Spectral leakage occurs when the record does not contain an integer number of cycles; window functions reduce leakage but broaden spectral peaks. The frequency resolution is of order 1/T for record duration T. Zero-padding interpolates the displayed spectrum but does not create new physical resolution.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

12. Laplace transforms

The unilateral Laplace transform is F(s)=∫₀∞f(t)e^{−st}dt. It converts linear differential equations with initial conditions into algebraic equations in s. Useful identities include L{f'}=sF(s)−f(0), L{f''}=s²F(s)−sf(0)−f'(0), and L{e^{at}}=1/(s−a) for Re(s)>a. The region of convergence matters, particularly for signals and stability analysis.

Solve y'+2y=0, y(0)=3. Transforming gives sY−3+2Y=0, so Y=3/(s+2), and inverse transformation yields y=3e^{−2t}. For y''+3y'+2y=0 with y(0)=1 and y'(0)=0, (s²Y−s)+3(sY−1)+2Y=0, hence Y=(s+3)/[(s+1)(s+2)]=2/(s+1)−1/(s+2), giving y=2e^{−t}−e^{−2t}.

Laplace transforms are especially effective for switched circuits, impulsive forces and initial-value problems. Partial fractions should be shown explicitly before inversion. For repeated poles, inverse transforms contain powers of t multiplying exponentials. A transform solution should be checked against the original differential equation and initial conditions.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

13. Ordinary differential equations and physical models

An ordinary differential equation relates a function of one independent variable to its derivatives. First-order equations may be separable, linear or exact; second-order linear equations with constant coefficients are solved using a characteristic polynomial. The nature of roots determines exponential, oscillatory or critically damped behaviour. Initial conditions select a particular solution, while boundary conditions constrain values at different positions.

For y''+4y=0, the characteristic equation r²+4=0 gives r=±2i. Therefore y=A cos2x+B sin2x. If y(0)=1 and y'(0)=0, then A=1 and B=0, so y=cos2x. For a damped oscillator y''+2βy'+ω₀²y=0, compare β with ω₀ to identify underdamped, critically damped and overdamped regimes.

Dimensional analysis checks the consistency of a proposed differential equation. For a mass-spring system, m x''+c x'+kx=F(t); m has units kg, c kg s⁻¹ and k kg s⁻². Each term has force units. Nondimensionalisation reveals controlling parameters and can simplify asymptotic analysis.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

14. Partial differential equations and separation

A partial differential equation contains derivatives with respect to more than one independent variable. The diffusion equation ∂T/∂t=α∇²T is parabolic; the wave equation ∂²u/∂t²=c²∇²u is hyperbolic; Laplace's equation is elliptic. These classifications describe how information and boundary data influence solutions. Separation of variables assumes a product form, such as u(x,t)=X(x)T(t), and converts a PDE into coupled ODEs linked by a separation constant.

For the one-dimensional wave equation u_tt=c²u_xx, let u=X(x)T(t). Then T''/(c²T)=X''/X=−k². Thus X''+k²X=0 and T''+c²k²T=0. Fixed ends X(0)=X(L)=0 give k_n=nπ/L, so the mode frequencies are ω_n=cnπ/L. Boundary conditions quantise the allowed modes.

A general solution is a sum of normal modes, with coefficients fixed by the initial displacement and velocity. Orthogonality is used to calculate those coefficients. Always check whether the assumed separated form satisfies both the PDE and every boundary condition.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

15. Heat equation and diffusion

The one-dimensional heat equation is T_t=αT_xx, where α is thermal diffusivity. For a rod of length L with fixed-temperature ends, separation gives sine eigenfunctions and exponentially decaying time factors. Higher spatial modes decay faster because their decay rate is proportional to n². Diffusion smooths spatial variation and has no finite propagation speed in the classical model.

For T_t=αT_xx with T(0,t)=T(L,t)=0, a separated mode is T_n(x,t)=A_n sin(nπx/L) exp[−α(nπ/L)²t]. If α=1.0×10⁻⁵ m² s⁻¹ and L=0.50 m, the fundamental decay constant is α(π/L)²=3.95×10⁻⁴ s⁻¹ and its e-folding time is about 2530 s.

The initial temperature profile is expanded as a Fourier sine series. Thermal diffusivity can be inferred from transient measurements by fitting mode decay rates, but boundary heat losses, sensor response and nonuniform material properties can invalidate the ideal model.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

16. Wave equation and normal modes

The wave equation describes propagation in strings, acoustic media and electromagnetic systems under appropriate approximations. For a stretched string, transverse displacement y(x,t) satisfies y_tt=(T/μ)y_xx, where T is tension and μ is mass per unit length. Fixed boundaries produce standing waves. Normal modes are independent patterns that oscillate at discrete frequencies determined by geometry and material properties.

For a string of length 0.80 m, tension 50 N and linear density 0.005 kg m⁻¹, wave speed c=√(T/μ)=√(10000)=100 m s⁻¹. The fundamental frequency is f₁=c/(2L)=62.5 Hz; the third harmonic is 187.5 Hz. This ideal result neglects bending stiffness, damping and frequency-dependent boundary effects.

A measured spectrum can identify resonances, but resonance peaks are affected by damping and excitation coupling. The mode shape can be measured using scanning sensors or visualisation. Compare measured frequencies with theory and discuss deviations rather than adjusting parameters without physical justification.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

17. Special functions: Bessel, Legendre and Hermite

Special functions arise when differential equations are separated in geometries with symmetry. Bessel functions J_m(kr) occur in cylindrical problems; Legendre polynomials P_l(cosθ) occur in spherical problems; Hermite polynomials occur in the quantum harmonic oscillator. Their zeros and orthogonality relations determine eigenvalues and expansion coefficients. These functions are not arbitrary formula collections: they encode boundary geometry and regularity at coordinate singularities.

The Bessel equation x²y''+xy'+(x²−m²)y=0 has regular solution J_m(x). For a circular membrane of radius a with fixed rim, radial modes require J_m(ka)=0. If j_{m,n} is the nth zero, then k_{m,n}=j_{m,n}/a. Frequencies therefore scale inversely with membrane radius for fixed wave speed.

Legendre polynomials begin P₀=1, P₁=x, P₂=(3x²−1)/2. They satisfy orthogonality on [−1,1]: ∫P_lP_m dx=2δ_lm/(2l+1). Expansion in these functions is useful for angular distributions and multipole solutions of Laplace's equation.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

18. Green functions and forced systems

A Green function G(x,ξ) is the response of a linear differential operator to a point source at ξ. If L y(x)=f(x), then y(x)=∫G(x,ξ)f(ξ)dξ, subject to compatible boundary conditions. The Green function satisfies L G=δ(x−ξ), with boundary conditions imposed in the observation coordinate. It converts a differential equation into an integral operation and makes the response to arbitrary forcing explicit.

For the operator −d²/dx² on 0<x<L with y(0)=y(L)=0, the Green function is G(x,ξ)=x(L−ξ)/L for x≤ξ and ξ(L−x)/L for x≥ξ. Its derivative has the jump required by the delta source. For constant forcing f₀, y(x)=∫₀ᴸG(x,ξ)f₀dξ=f₀x(L−x)/2, which satisfies −y''=f₀ and both boundary conditions.

Green functions are used in electrostatics, elasticity, wave propagation and quantum scattering. Check operator sign conventions: changing L to −L changes the Green function sign. Also verify continuity and derivative jump conditions at the source point.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

19. Calculus of variations and Euler–Lagrange equation

The calculus of variations finds functions that make a functional stationary. For J[y]=∫_{x₁}^{x₂}F(x,y,y')dx with fixed endpoints, a small variation y→y+εη leads, after integration by parts, to the Euler–Lagrange equation ∂F/∂y−d/dx(∂F/∂y')=0. In mechanics, the stationary action S=∫L(q,q̇,t)dt yields Lagrange's equations.

For the shortest path between two points in a plane, arc length is J[y]=∫√(1+y'²)dx. Since F is independent of y, Euler–Lagrange gives d/dx[y'/√(1+y'²)]=0. Hence y' is constant and the extremal is a straight line. This derives a familiar geometric result from a variational principle.

A stationary path need not always be a minimum; it may be a maximum or saddle. Physical action principles often produce stationary action rather than globally minimal action. State endpoint conditions and any constraints before applying the equation.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

20. Numerical methods and error control

Many physical equations have no closed-form solution. Numerical methods approximate roots, integrals, derivatives and differential equations. Bisection is robust for bracketed roots; Newton's method is fast near a simple root but depends on a good initial estimate. Trapezoidal and Simpson rules approximate integrals; Euler, Runge–Kutta and adaptive methods solve initial-value ODEs. Discretisation, round-off, truncation and model errors are distinct.

For f(x)=x²−2, Newton iteration is x_{n+1}=x_n−(x_n²−2)/(2x_n). Starting x₀=1.5 gives x₁=1.4166667, x₂≈1.4142157 and x₃≈1.41421356. Check convergence with both step size and residual. A small change between iterates alone may be misleading if the function is poorly scaled.

Convergence should be tested by refining step size or grid spacing and comparing results. A method's formal order predicts how error scales asymptotically, but only when the solution is sufficiently smooth and the numerical regime is appropriate. Preserve significant digits internally and round only for reporting.

Problem-solving procedure. Identify the unknown function or quantity and the independent variable. Write down the mathematical operator and its domain. For differential equations, state initial or boundary conditions; for transforms, state the transform convention and convergence assumptions; for vector integrals, state orientation and geometry. Carry constants symbolically until the structure of the solution is clear. This reduces algebraic errors and makes the dependence on physical parameters visible.

Verification and interpretation. Substitute the proposed result into the original equation rather than relying only on the steps used to obtain it. Check boundary values, initial values, dimensions, signs, limiting behaviour and special cases. If a numerical approximation is used, vary the step size or tolerance and inspect the residual. A mathematically correct expression can still be physically inappropriate if it violates the domain, boundary conditions or assumptions of the model.

Figure 2. Fourier decomposition into harmonics
fundamental and higher modes
Figure 3. Separation of variables for a vibrating string
fixed end x=0fixed end x=Lmode shape Xₙ(x)

30 worked calculations with complete method

These examples deliberately cover different mathematical tools. In an examination, write each algebraic step and state the assumptions rather than jumping directly to the answer.

Worked calculation 1: Complex multiplication

Compute (3+2i)(1−4i). Expand: 3−12i+2i−8i²=11−10i. Modulus of the result is √(121+100)=√221≈14.866.

Worked calculation 2: Complex division

Evaluate (2+3i)/(1−i). Multiply numerator and denominator by 1+i: [(2+3i)(1+i)]/2=(−1+5i)/2=−0.5+2.5i.

Worked calculation 3: Polar form

For z=−1+√3i, r=√(1+3)=2 and θ=2π/3. Thus z=2e^{i2π/3}. Then z³=8e^{i2π}=8.

Worked calculation 4: Roots of a complex number

Find square roots of 4i. Write 4i=4e^{iπ/2}. Roots are 2e^{iπ/4} and 2e^{i5π/4}, namely √2(1+i) and −√2(1+i). Squaring either gives 4i.

Worked calculation 5: Cauchy–Riemann test

For f(z)=z², u=x²−y² and v=2xy. u_x=2x=v_y and u_y=−2y=−v_x, so f is analytic and f'=2z.

Worked calculation 6: Taylor expansion

Expand e^{2x} to x³: 1+2x+(2x)²/2!+(2x)³/3! =1+2x+2x²+(4/3)x³+O(x⁴). At x=0.1 this gives 1.221333, compared with exact 1.221403.

Worked calculation 7: Matrix solution

Solve 3x+2y=12 and x−y=1. From x=1+y, substitute: 3+3y+2y=12, y=1.8 and x=2.8. Check 3(2.8)+2(1.8)=12.

Worked calculation 8: Determinant and inverse

For A=[[4,1],[2,3]], det=12−2=10 and A⁻¹=(1/10)[[3,−1],[−2,4]]. The nonzero determinant confirms a unique solution for every right-hand side.

Worked calculation 9: Eigenvalues

For A=[[5,2],[2,5]], characteristic equation (5−λ)²−4=0 gives λ=7,3. Eigenvectors are (1,1) and (1,−1), respectively.

Worked calculation 10: Gradient

If V=4x²+2xy−z, then ∇V=(8x+2y,2x,−1). At (1,2,3), ∇V=(12,2,−1). Electric field E=−∇V=(-12,-2,1) in the corresponding units.

Worked calculation 11: Divergence

For F=(x²y, yz², 3z), ∇·F=2xy+z²+3. At (1,2,2), divergence=4+4+3=11.

Worked calculation 12: Curl

For F=(−y/2,x/2,0), curl F=(0,0,1). The field has uniform positive z-directed local circulation.

Worked calculation 13: Line integral

For F=(2x,2y), integrate from (0,0) to (1,2) along any path. Since F=∇(x²+y²), work=1²+2²−0=5.

Worked calculation 14: Flux through a sphere

For F=r⃗=(x,y,z), divergence=3. Through sphere radius R, flux=∭3dV=3(4πR³/3)=4πR³. At R=2, flux=32π.

Worked calculation 15: Laplacian

For φ=x²+y²−2z², ∇²φ=2+2−4=0. It is harmonic in all space.

Worked calculation 16: Fourier coefficient

For f(x)=x on (−π,π), b₁=(1/π)∫x sinx dx=2. The first harmonic approximation is 2sinx; adding terms improves the representation away from the periodic jump.

Worked calculation 17: Fourier series value

Use x=2(sin x−sin2x/2+sin3x/3−…). At x=π/2, terms are 2(1−0−1/3+0+1/5−…); the series converges to π/2.

Worked calculation 18: Fourier transform

For e^{−a|x|}, transform is 2a/(a²+k²). With a=2 and k=1, F(1)=4/5=0.8.

Worked calculation 19: Laplace transform

L{t²}=2/s³. This follows by two integrations by parts or the general result L{tⁿ}=n!/sⁿ⁺¹ for Re(s)>0.

Worked calculation 20: Inverse Laplace

Y(s)=5/[s(s+2)]. Partial fractions give (5/2)(1/s−1/(s+2)); y(t)=(5/2)(1−e^{−2t}).

Worked calculation 21: Second-order ODE

Solve y''−5y'+6y=0. Characteristic roots r=2,3, so y=Ae^{2x}+Be^{3x}. For y(0)=1,y'(0)=0, A=3 and B=−2.

Worked calculation 22: Damped oscillator

For x''+4x'+13x=0, roots are −2±3i. Thus x=e^{−2t}(Acos3t+Bsin3t), an underdamped response.

Worked calculation 23: Wave speed

A string has tension 80 N and μ=0.020 kg/m. c=√(80/0.020)=√4000=63.25 m/s.

Worked calculation 24: String fundamental

For L=1.2 m and c=63.25 m/s, f₁=c/(2L)=26.35 Hz; f₄=105.4 Hz.

Worked calculation 25: Heat diffusion time

For α=8×10⁻⁶ m²/s and L=0.20 m, fundamental decay rate α(π/L)²=1.97×10⁻³ s⁻¹; τ≈508 s.

Worked calculation 26: Green function solution

For −y''=2 on 0<x<1 with y(0)=y(1)=0, y=x(1−x). Directly, y''=−2, so −y''=2.

Worked calculation 27: Newton root

For x²−3=0 and x₀=2, x₁=2−1/8=1.875; x₂≈1.820833; further iterations approach √3≈1.73205.

Worked calculation 28: Simpson integration

Approximate ∫₀²x²dx with Simpson's rule using h=1: (h/3)[f(0)+4f(1)+f(2)]=(1/3)(0+4+4)=8/3, exact for a quadratic.

Worked calculation 29: Dimensional check

For ω=√(k/m), k has units N/m=kg s⁻²; k/m has units s⁻², so ω has units s⁻¹ as required.

Worked calculation 30: Variational shortest path

For F=√(1+y'²), ∂F/∂y=0 and d/dx[y'/√(1+y'²)]=0. Hence y'=constant and y=mx+c, a straight line.

Additional derivations and mathematical checks

Derivation A: Fourier sine coefficients

For an odd function on (−L,L), cosine coefficients vanish. The sine coefficients are b_n=(1/L)∫_{−L}^{L}f(x)sin(nπx/L)dx=(2/L)∫₀ᴸf(x)sin(nπx/L)dx. For f(x)=x and L=π, integration by parts gives b_n=(2/π)[−x cos(nx)/n+sin(nx)/n²]₀^π=2(−1)^{n+1}/n. Therefore x=2Σ(−1)^{n+1}sin(nx)/n for −π

Derivation B: Laplace transform of a second derivative

By definition L{y''}=∫₀∞y''(t)e^{−st}dt. Integrating by parts gives [y'e^{−st}]₀∞+s∫₀∞y'e^{−st}dt=−y'(0)+sL{y'}. Applying integration by parts again to L{y'} gives sY−y(0). Hence L{y''}=s²Y−sy(0)−y'(0). The initial values enter as algebraic terms, which is why the method is efficient for initial-value problems.

Derivation C: Normal-mode frequencies of a fixed string

For y_tt=c²y_xx, assume y=XT. Then T''/c²T=X''/X=−k². The spatial solution is X=A sin(kx)+B cos(kx). The condition X(0)=0 gives B=0. X(L)=0 gives sin(kL)=0, so kL=nπ and k_n=nπ/L. The time equation is T''+c²k_n²T=0, hence ω_n=cnπ/L and f_n=cn/(2L). The boundary conditions produce the discrete harmonic spectrum.

Derivation D: Green-function jump condition

For −d²G/dx²=δ(x−ξ), integrate over (ξ−ε,ξ+ε): −[G'(ξ+ε)−G'(ξ−ε)]=1 as ε→0. Thus G is continuous at ξ but its derivative has jump G'(ξ⁺)−G'(ξ⁻)=−1. This jump is the mathematical signature of the point source. Applying the boundary conditions determines the two linear pieces of G.

Derivation E: Newton iteration

For f(x)=0, approximate f near x_n by f(x_n)+f'(x_n)(x−x_n). Set this linear approximation to zero and solve for x: x_{n+1}=x_n−f(x_n)/f'(x_n). The method converges quadratically near a simple root when the initial estimate is sufficiently close and the derivative is nonzero. Multiple roots generally reduce the convergence order.

Mathematical modelling in physics: worked application

A damped, driven oscillator obeys m x''+c x'+kx=F₀cos(ωt). Divide by m to obtain x''+2βx'+ω₀²x=(F₀/m)cos(ωt), where β=c/(2m) and ω₀=√(k/m). The transient solution is determined by the roots of r²+2βr+ω₀²=0. For the underdamped case, x_h=e^{−βt}[Acos(ω_dt)+Bsin(ω_dt)], with ω_d=√(ω₀²−β²). For steady-state motion, assume x_p=Xcos(ωt−δ). Substitution gives amplitude X=(F₀/m)/√[(ω₀²−ω²)²+(2βω)²] and phase tanδ=2βω/(ω₀²−ω²), with quadrant chosen consistently. This single model uses differential equations, complex amplitudes, algebra and limiting-case analysis.

For m=0.50 kg, k=200 N m⁻¹ and c=2.0 kg s⁻¹, ω₀=√(200/0.50)=20 rad s⁻¹ and β=2/(2×0.50)=2 s⁻¹. Thus ω_d=√(400−4)=19.90 rad s⁻¹. The transient amplitude envelope decays as e^{−2t}; its time constant is 1/β=0.50 s. These values should be checked dimensionally: β and ω have units s⁻¹, while the exponent βt is dimensionless.

University examination questions with model answers

1. State Euler's identity and explain its use in oscillatory physics.

Model answer: e^{iθ}=cosθ+i sinθ. It represents sinusoidal signals compactly and converts differentiation into multiplication by iω.

2. What conditions must u and v satisfy for f=u+iv to be analytic?

Model answer: The Cauchy–Riemann equations u_x=v_y and u_y=−v_x, together with suitable continuity of first partial derivatives in a neighbourhood.

3. Distinguish Taylor and Laurent series.

Model answer: Taylor series has nonnegative powers about a regular point; Laurent series may include negative powers and describes functions in an annulus around an isolated singularity.

4. What does a nonzero determinant imply?

Model answer: The square matrix is invertible and the associated linear system has a unique solution for every right-hand side.

5. Define an eigenvalue and eigenvector.

Model answer: A scalar λ and nonzero vector v satisfying Av=λv.

6. Why are eigenvectors useful in coupled oscillations?

Model answer: They give independent normal-mode shapes that diagonalise the linear equations; eigenvalues determine modal frequencies or growth rates.

7. Distinguish dot and cross products.

Model answer: Dot product is scalar and measures projection; cross product is a vector normal to the two vectors with magnitude ab sinθ.

8. Give physical meanings of gradient, divergence and curl.

Model answer: Gradient gives steepest scalar increase; divergence measures local source density; curl measures local circulation.

9. State the divergence theorem.

Model answer: The outward flux through a closed surface equals the volume integral of the divergence within it.

10. State Stokes' theorem.

Model answer: The circulation around a closed curve equals the surface integral of the curl through an oriented surface bounded by that curve.

11. What is a harmonic function?

Model answer: A twice-differentiable scalar function satisfying Laplace's equation ∇²φ=0.

12. Explain Fourier orthogonality.

Model answer: Different sine and cosine modes integrate to zero over a full period under the appropriate weighting, allowing independent coefficient extraction.

13. What is Gibbs phenomenon?

Model answer: Persistent overshoot near a jump discontinuity in Fourier partial sums, whose width shrinks but peak overshoot does not vanish in the usual limit.

14. State the convolution theorem.

Model answer: Under a consistent transform convention, convolution in the original domain corresponds to multiplication in transform space.

15. Why are Laplace transforms useful for initial-value problems?

Model answer: They transform derivatives into algebraic expressions that explicitly include initial values.

16. Classify diffusion, wave and Laplace equations.

Model answer: Diffusion is parabolic, wave is hyperbolic, and Laplace is elliptic.

17. What determines allowed normal-mode frequencies?

Model answer: The differential operator, material parameters, geometry and boundary conditions.

18. Name three special functions and their common geometries.

Model answer: Bessel functions for cylindrical symmetry, Legendre functions for spherical angular dependence, and Hermite functions for the harmonic oscillator.

19. Define a Green function.

Model answer: The response of a linear differential operator to a point-source delta function, subject to specified boundary conditions.

20. Write the Euler–Lagrange equation.

Model answer: ∂F/∂y−d/dx(∂F/∂y')=0 for a functional ∫F(x,y,y')dx with fixed endpoints.

21. Compare bisection and Newton methods.

Model answer: Bisection is robust when a sign-changing bracket exists but converges linearly; Newton is often faster near a simple root but needs a good initial guess and derivative.

22. What is numerical truncation error?

Model answer: The error introduced by replacing an exact mathematical operation or infinite process with a finite approximation.

23. Why refine a numerical grid?

Model answer: To assess discretisation convergence and estimate whether the computed solution is approaching a stable continuum result.

24. What is a condition number?

Model answer: A measure of sensitivity of a problem's solution to small perturbations in input data.

25. How do you verify a derived mathematical solution?

Model answer: Substitute it into the governing equation, test boundary and initial conditions, check dimensions, limiting cases and physical plausibility.

Formula sheet and revision checklist

z=re^{iθ}; e^{iθ}=cosθ+i sinθ; det(A−λI)=0; ∇²φ=∇·∇φ

Fourier series: f(x)=a₀/2+Σ[a_n cos(nπx/L)+b_n sin(nπx/L)]

Fourier transform: F(k)=∫f(x)e^{−ikx}dx; inverse f(x)=(1/2π)∫F(k)e^{ikx}dk

Laplace: L{f'}=sF−f(0); L{f''}=s²F−sf(0)−f'(0)

Euler–Lagrange: ∂F/∂y−d/dx(∂F/∂y')=0

Wave speed string: c=√(T/μ); fixed-string modes: f_n=nc/(2L)

Extended calculation workshop: full derivations and examination-style problems

1. Complex impedance and phase angle

A series circuit contains R=30 Ω and an inductance L=0.20 H at angular frequency ω=100 rad s⁻¹. The inductive reactance is X_L=ωL=100×0.20=20 Ω. The impedance is Z=R+iX_L=30+20i Ω. Its magnitude is |Z|=√(30²+20²)=√1300=36.06 Ω and its phase is θ=tan⁻¹(20/30)=33.69°. If the applied RMS voltage is 120 V, the RMS current magnitude is I=V/|Z|=120/36.06=3.328 A. The current lags the voltage by 33.69° because the circuit is inductive. The real power is I²R=(3.328)²×30≈332 W, while apparent power is VI≈399 VA. Check that the power factor is cosθ=30/36.06=0.832 and 399×0.832≈332 W. This example combines complex numbers, phasors and physical interpretation.

2. Diagonalising a coupled two-degree-of-freedom system

Consider two identical masses coupled symmetrically with mass-normalised stiffness matrix K=[[3,−1],[−1,3]] in consistent units. The eigenvalue equation is det(K−λI)=(3−λ)²−1=0. Thus 3−λ=±1, giving λ₁=2 and λ₂=4. The eigenvector for λ₁ satisfies x−y=0, so the normal coordinate is proportional to (1,1); for λ₂, x+y=0, giving (1,−1). If λ represents ω², the angular frequencies are √2 and 2 in the associated units. The first mode has the masses moving together; the second has them moving oppositely. The transformation to normal coordinates diagonalises the quadratic energy and separates the equations. In an actual mechanical system, construct K from the spring potential energy and M from kinetic energy, then solve the generalised eigenproblem Kv=ω²Mv.

3. Green function for a one-dimensional boundary problem

Solve −y''=f(x) on 0<x<L with y(0)=y(L)=0. The Green function is constructed from two homogeneous solutions: one satisfying the left boundary and one the right boundary. Choose u₁=x and u₂=L−x. Their Wronskian u₁u₂'−u₁'u₂=−L. The resulting kernel is G(x,ξ)=x(L−ξ)/L for x≤ξ and ξ(L−x)/L for x≥ξ. For constant f(x)=f₀, split the integral at ξ=x: y(x)=∫₀ˣ[ξ(L−x)/L]f₀dξ+∫ₓᴸ[x(L−ξ)/L]f₀dξ. The first term is f₀(L−x)x²/(2L); the second is f₀x(L−x)²/(2L). Adding gives f₀x(L−x)/2. Differentiating twice yields y''=−f₀, and y vanishes at both endpoints. The direct substitution is an essential verification.

4. Fourier transform of a Gaussian

Consider f(x)=exp(−ax²), where a>0. Its Fourier transform under F(k)=∫exp(−ax²)exp(−ikx)dx is √(π/a)exp[−k²/(4a)]. One route is to complete the square in the exponent, −ax²−ikx=−a[x+ik/(2a)]²−k²/(4a), and use the Gaussian integral ∫e^(−ax²)dx=√(π/a). The result is another Gaussian: a narrow function in x has a broad transform in k. For a=4, F(0)=√π/2≈0.8862 and F(2)=0.8862e^(−1/4)≈0.6904. The transform convention determines any factors of 2π.

5. Laplace transform solution of a driven first-order circuit

A series RC circuit is connected to a DC source V₀ at t=0. The capacitor voltage obeys RC dv_C/dt+v_C=V₀, with v_C(0)=0. Let τ=RC. Transforming gives τ[sV(s)−0]+V(s)=V₀/s. Hence V(s)=V₀/[s(τs+1)]. Partial fractions give V(s)=V₀[1/s−1/(s+1/τ)]. Inverting yields v_C(t)=V₀[1−e^(−t/τ)]. At t=τ, v_C=0.632V₀; at t=3τ, v_C≈0.950V₀; at t=5τ, v_C≈0.993V₀. This gives the time constant a direct operational meaning. The current is i(t)=(V₀/R)e^(−t/τ), and integrating current over time gives total capacitor charge C V₀.

6. Separation of variables for a finite heat-conduction problem

For a rod with zero temperature excess at x=0 and x=L, solve T_t=αT_xx with initial profile T(x,0)=T₀ sin(πx/L). Separation gives eigenmodes sin(nπx/L)e^(−αn²π²t/L²). Since the initial condition is exactly the fundamental mode, all higher coefficients vanish: T(x,t)=T₀ sin(πx/L)e^(−απ²t/L²). At the rod centre x=L/2, the sine factor is one. If α=1.2×10⁻⁵ m² s⁻¹ and L=0.30 m, decay rate is α π²/L²≈1.316×10⁻³ s⁻¹. After 300 s, T/T₀=e^(−0.3948)≈0.674. The time dependence and spatial profile can be checked independently in the PDE.

7. Spherical symmetry and the radial Laplacian

For a radially symmetric scalar φ(r) in three dimensions, ∇²φ=(1/r²)d/dr[r²dφ/dr]. In a source-free spherical region, set this equal to zero. Integrate once: r²φ'=C₁, so φ'=C₁/r². Integrate again: φ=−C₁/r+C₂. Thus the general spherically symmetric harmonic function is A+B/r. The 1/r dependence appears in electrostatic potential outside a point charge and gravitational potential outside a spherical mass. The solution is singular at r=0 unless B=0; whether the origin belongs to the physical domain determines whether that term is admissible.

8. Residue calculation of a real integral

Consider the integral over the real line of 1/(x²+a²), with a>0. Extend the function to z and integrate over a semicircular contour in the upper half-plane. The only enclosed pole is z=ia, with residue lim(z→ia)(z−ia)/(z²+a²)=1/(2ia). The residue theorem gives contour integral 2πi/(2ia)=π/a. The large semicircle contribution tends to zero because the integrand scales as 1/R² while arc length scales as R. Therefore ∫₋∞∞dx/(x²+a²)=π/a. By evenness, ∫₀∞dx/(x²+a²)=π/(2a). The contour choice and decay estimate are both necessary parts of a rigorous argument.

9. Numerical integration and convergence check

Approximate ∫₀¹e^(−x²)dx. With the trapezoidal rule and one interval, T₁=[f(0)+f(1)]/2=(1+e⁻¹)/2≈0.68394. With two intervals h=0.5, T₂=0.25[1+2e⁻⁰·²⁵+e⁻¹]≈0.73137. Simpson's rule with two intervals gives S₂=(h/3)[f(0)+4f(0.5)+f(1)]≈0.74718. A high-accuracy value is approximately 0.746824. Simpson's rule is exact for polynomials through degree three and has higher order than the trapezoidal rule for smooth functions. Refinement and comparison with a trusted reference help estimate discretisation error.

10. Dimensional analysis and nondimensionalisation

For a mass-spring-damper equation m x''+c x'+kx=0, choose a characteristic time t₀=√(m/k) and set τ=t/t₀. Then x'=x_τ/t₀ and x''=x_ττ/t₀². Substitution gives x_ττ+[c/√(mk)]x_τ+x=0. The dimensionless damping parameter is c/√(mk), equal to twice the damping ratio ζ. This exposes the true independent control parameter and allows different physical systems with the same ζ to share the same dimensionless response. Nondimensionalisation also helps detect missing factors and makes numerical values more transferable.

11. Error propagation in a derived physical quantity

Suppose wave speed is calculated from c=√(T/μ), where tension T and linear density μ are independent. Taking logarithms gives ln c=(1/2)ln T−(1/2)ln μ. Differentiation gives dc/c=(1/2)dT/T−(1/2)dμ/μ. For independent standard uncertainties, [u(c)/c]²=(1/4)[u(T)/T]²+(1/4)[u(μ)/μ]². If T=80.0±0.8 N and μ=0.0200±0.0004 kg m⁻¹, relative uncertainties are 1.0% and 2.0%. Thus relative uncertainty in c is 0.5√(0.01²+0.02²)=1.118%. Since c=63.25 m s⁻¹, u(c)≈0.71 m s⁻¹.

12. Choosing and checking a boundary condition

For a string fixed at both ends, displacement is zero at x=0 and x=L. For a string with one free end, the transverse force condition commonly gives zero slope at that end in the ideal model. These are physically different eigenvalue problems and produce different allowed wavenumbers. For a fixed-free string, X(0)=0 and X'(L)=0 give cos(kL)=0, so k_n=(n+1/2)π/L for n=0,1,2,…; the fundamental frequency is c/(4L), half the fixed-fixed fundamental. A mathematical solution that satisfies the PDE but the wrong end condition is not a solution to the physical experiment.

Further advanced worked problems

13. Contour integration for a second rational integral

Evaluate ∫₋∞∞ dx/(x²+a²)² for a>0. Use a semicircular contour in the upper half-plane. There is a double pole at z=ia. Write f(z)=1/[(z−ia)²(z+ia)²]. The residue at the double pole is the derivative of 1/(z+ia)²: Res=factor derivative at z=ia, equal to −2/(2ia)³=1/(4ia³). The residue theorem gives 2πi×1/(4ia³)=π/(2a³). The arc contribution vanishes because the integrand is O(R⁻⁴) and arc length is O(R). Therefore the real integral is π/(2a³). This result can be checked by differentiating the known integral with respect to a: d/da ∫dx/(x²+a²)=−2a∫dx/(x²+a²)², giving the same answer. The independent check is valuable because double-pole residue signs are a common source of errors.

14. Fourier coefficients for a constant-plus-ramp profile

Let f(x)=C+x on −L<x<L, periodically extended. The constant C contributes only to a₀: a₀=(1/L)∫₋ᴸᴸ(C+x)dx=2C. The odd x term contributes no cosine coefficients. Thus a_n=0 for n≥1. The sine coefficient is b_n=(1/L)∫₋ᴸᴸx sin(nπx/L)dx=(2/L)∫₀ᴸx sin(nπx/L)dx. Integration by parts gives b_n=2L(−1)^{n+1}/(nπ). The full expansion is C+2L/π[sin(πx/L)−sin(2πx/L)/2+sin(3πx/L)/3−…]. At the discontinuity created by periodic extension, the series converges to the average of the left and right limits, not to an arbitrarily assigned endpoint value.

15. Eigenvalue calculation for a three-point finite-difference model

Discretise −u''=λu on 0<x<1 with u(0)=u(1)=0 using three equally spaced interior nodes and step h=1/4. The central difference gives −(u_{j+1}−2u_j+u_{j−1})/h²=λu_j. The matrix is (1/h²)[[2,−1,0],[−1,2,−1],[0,−1,2]]. Its eigenvalues are λ_k=(2/h²)[1−cos(kπ/4)] for k=1,2,3. Since 1/h²=16, λ₁=32(1−√2/2)≈9.373, λ₂=32, and λ₃=32(1+√2/2)≈54.627. The continuum eigenvalues are (kπ)²: approximately 9.870, 39.478 and 88.826. The coarse grid underestimates the first eigenvalue and poorly resolves higher modes. Refining the grid improves the low-mode approximation and is essential for convergence studies.

16. Coupled first-order equations and matrix exponentials

Consider dX/dt=AX with A=[ [0,1], [−4,−3] ]. The characteristic polynomial is λ²+3λ+4=0, with roots λ=−1.5±i√7/2. The solution has damped oscillatory components proportional to e^(−1.5t)cos(√7 t/2) and e^(−1.5t)sin(√7 t/2). This follows because the matrix exponential e^(At) is the fundamental solution. For an initial vector X(0), X(t)=e^(At)X(0). Eigenvalues determine stability: negative real parts imply decay, positive real parts imply growth, and zero real parts require further analysis. The result connects linear algebra to coupled dynamical systems.

17. A spherical electrostatic potential from Laplace's equation

In a charge-free spherical region with spherical symmetry, ∇²V=0 reduces to (1/r²)d/dr(r²dV/dr)=0. Integrating gives r²V'=C₁, then V=−C₁/r+C₂. If V→0 as r→∞, then C₂=0. If the potential at radius R is V₀, then V(R)=−C₁/R=V₀, so C₁=−V₀R and V(r)=V₀R/r for r≥R. The electric field is E_r=−dV/dr=V₀R/r². At r=2R, the potential is V₀/2 and field magnitude is V₀/(4R). This is the mathematical form of the exterior potential of a spherically symmetric charged conductor.

18. Runge–Kutta step for exponential decay

Solve y'=−2y, y(0)=1 for one step h=0.1 using classical fourth-order Runge–Kutta. k₁=−2. k₂=−2(1+0.05k₁)=−1.8. k₃=−2(1+0.05k₂)=−1.82. k₄=−2(1+0.1k₃)=−1.636. Then y₁=1+(0.1/6)(k₁+2k₂+2k₃+k₄)=1−0.1812667=0.8187333. The exact value is e⁻⁰·²=0.8187308, so the one-step error is about 2.5×10⁻⁶. Repeat with h=0.05 and compare errors; for a fourth-order method, the global error should decrease approximately by a factor of 16 when the step is halved in the asymptotic regime.

Calculation strategy, checks and common sources of error

Choosing the mathematical representation

Before starting an extended calculation, decide whether the problem is best represented by a scalar equation, vector equation, matrix system, transform or differential operator. A periodic steady-state circuit is naturally treated with phasors; a transient circuit with specified initial charge is often simpler with Laplace transforms. A symmetric field suggests a coordinate system aligned with that symmetry. A set of coupled linear oscillators suggests a matrix eigenvalue problem. The correct representation reduces algebra and exposes the physical structure.

Boundary conditions are part of the problem

A differential equation alone usually has infinitely many solutions. Initial conditions specify the state at one time; boundary conditions specify values or derivatives at spatial boundaries. Dirichlet conditions prescribe the function, Neumann conditions prescribe its normal derivative, and Robin conditions prescribe a linear combination. For heat flow, an insulated boundary has zero normal temperature gradient; a fixed-temperature boundary has prescribed temperature. For a string, fixed ends have zero displacement. Selecting the wrong condition changes the eigenvalues and therefore the predicted physical frequencies or decay rates.

Dimensional and limiting checks

Check every term in an equation has the same dimensions. Exponents and trigonometric arguments must be dimensionless. A frequency must have units s⁻¹, a potential must have the appropriate energy per charge, and a diffusion exponent αk²t must be dimensionless. Then test limiting cases: zero damping should recover undamped oscillation; zero forcing should remove the steady forced response; large time in a stable diffusion mode should approach equilibrium; and a vanishing source should recover the homogeneous equation. These checks catch errors that can survive several lines of algebra.

Numerical stability and reporting

Distinguish a converged numerical answer from a merely plausible decimal. State the method, step size, stopping tolerance, initial estimate, boundary treatment and convergence test. Report enough digits to show numerical stability but do not imply more physical precision than the input data justify. When a method fails, investigate ill-conditioning, stiffness, singularities, discontinuities or an inappropriate step size. A graph is useful, but it does not replace a residual, error estimate or analytical check.

Common examination errors

Additional derivation practice: from equation to physical result

Orthogonality of sine modes

For positive integers m and n, consider I=∫₀ᴸsin(mπx/L)sin(nπx/L)dx. Use the product identity sinA sinB=[cos(A−B)−cos(A+B)]/2. When m≠n, both cosine terms integrate to sine functions evaluated at integer multiples of π, so I=0. When m=n, the integrand is sin²(nπx/L)=[1−cos(2nπx/L)]/2, giving I=L/2. Thus the modes are orthogonal and their squared norm is L/2. This result is the foundation for extracting Fourier sine coefficients and for projecting an initial condition onto normal modes.

Deriving the resonance condition

For a driven damped oscillator, the steady amplitude denominator is D(ω)=(ω₀²−ω²)²+4β²ω². The amplitude is maximum where D is minimum. Differentiate: D'=−4ω(ω₀²−ω²)+8β²ω=4ω[−ω₀²+ω²+2β²]. Apart from ω=0, the stationary point satisfies ω²=ω₀²−2β². Therefore the displacement-amplitude resonance frequency is ω_r=√(ω₀²−2β²), provided ω₀²>2β². It is slightly below the undamped natural frequency. This is distinct from the damped free-oscillation frequency √(ω₀²−β²); confusing these two frequencies is a common conceptual error.

Checking a solution with a residual

Suppose an approximate numerical function y_h is proposed for y''+p(x)y'+q(x)y=f(x). Define residual R(x)=y_h''+p(x)y_h'+q(x)y_h−f(x). A small residual indicates that the differential equation is approximately satisfied at the evaluated points, but it does not alone guarantee a small solution error: boundary-condition errors and ill-conditioning can amplify the effect. Evaluate the residual over the whole domain, check boundary mismatches, and compare solutions on refined grids. For a linear problem, a posteriori error estimates can use stability bounds relating residual size to solution error.

Reporting advanced mathematical calculations

When presenting a final answer, identify whether a solution is exact, asymptotic or numerical. Give the valid domain and any excluded singular points. State whether constants are dimensional or dimensionless, and define every symbol at first use. For numerical tables, include units and sufficient precision to reproduce the calculation. For plots, label axes, indicate parameter values and distinguish computed data from analytical curves. If several mathematical branches exist, explain which branch is selected by the physical boundary or initial conditions. This reporting discipline makes a calculation independently checkable and prevents a correct intermediate expression from being misinterpreted as the final physical prediction.

Final physical checks for mathematical solutions

For a calculated field, inspect symmetry, continuity and behaviour at infinity or at the origin where relevant. For a wave solution, check propagation speed, phase direction and energy flow. For a diffusion solution, check that the profile smooths with time and that the boundary heat flux agrees with the imposed condition. For an eigenvalue problem, check orthogonality when the operator is self-adjoint and verify the eigenpair by direct substitution. For a transform solution, confirm the transform convention and region of convergence. These checks turn formal manipulation into a defensible physical calculation and should appear explicitly in advanced laboratory and examination work.

Further study

Unit 18: Computational Physics I applies algorithms and numerical analysis to physical problems.

Continue to Unit 18: Computational Physics I →

Official Education References

Use these official organisations for curriculum guidance, examinations and further learning information.