Hyperbolic partial differential equations describe fields whose information travels along characteristic directions, as in ideal wave propagation and many conservation laws. A dependable solution must identify those propagation speeds, supply the right initial and boundary data, and distinguish a smooth wave from a shock that requires an integral interpretation.
The gold standard is to make those distinctions calculable. This guide develops transport, vibrating strings, domains of dependence, nonlinear steepening, shocks and numerical fluxes through small worked examples. All numerical cases are constructed teaching models with explicit assumptions, not forecasts for a particular road, structure or fluid system.
Start with The Gold Standard master for the collection, the Education guide for learning that transfers, and How X Works for connections to mechanisms in other subjects.
Choose your hyperbolic PDE route
Chapters 1–3: Follow information and boundaries
Chapters 4–6: Calculate waves and dependence
Chapters 7–9: Connect energy and nonlinear motion
Chapters 10–12: Resolve shocks and computation
Chapter index
Follow information and boundaries · Chapters 1–3
1. Ask what moves and what carries the information
Calculate waves and dependence · Chapters 4–6
4. Derive the string wave speed from a force balance
Connect energy and nonlinear motion · Chapters 7–9
7. Check energy and reflection rather than expecting decay
Resolve shocks and computation · Chapters 10–12
10. Determine a shock speed from conservation
Transfer and continue · Chapters 13–14
13. Transfer the reasoning when the flux changes
14. Use a diagnostic and learning route that survives new examples
CHAPTER 1 OF 14 · Follow information and boundaries
1. Ask what moves and what carries the information
A field can change because its values are transported, because disturbances propagate through an interacting medium, or because sources alter the quantity locally. These mechanisms require different equations. Before classifying a PDE, identify the unknown, its units, the spatial region and the physical meaning of time.
The simplest transport equation is u_t + a u_x = 0, where a is a constant velocity. Its unknown might be a concentration carried without diffusion by a prescribed flow. The wave equation u_tt = c²u_xx instead describes an ideal disturbance with propagation speed c. For a string, u is transverse displacement rather than the amount of a conserved substance.
Both examples have characteristic directions, but they need different initial information. The transport equation is first order in time and needs one initial profile. The wave equation is second order in time and needs initial displacement and initial velocity. Copying the data structure of one into the other leaves either missing information or an inappropriate constraint.
Units expose a mistaken interpretation
If u is displacement in metres, u_tt has units metres per second squared. Since u_xx has units inverse metres, c² must have units square metres per square second. Thus c has units of speed. It is the speed of the disturbance along the string, not generally the transverse velocity of a piece of string, which is u_t.
The existing Partial Differential Equations guide provides the wider field-based viewpoint. Here the central question is more specific: which earlier data can influence the value at a chosen position and time, and what happens when different pieces of that information converge?
CHAPTER 2 OF 14 · Follow information and boundaries
2. Follow a transported profile along characteristics
For constant a, set x(t) = ξ + at, where ξ labels a starting position. Along that path, the chain rule gives d[u(x(t),t)]/dt = u_t + a u_x = 0. The value is constant along the path. If u(x,0) = g(x), the exact answer is u(x,t) = g(x − at).
The sign is worth checking. A feature initially at x = 3 moves to x = 3 + at. For a = 2 and t = 4, it reaches x = 11. Substituting x − at = 3 at that new position recovers the initial feature. Writing g(x + at) would move it the other way.
Translate a finite pulse
Take g(x) = 1 for 0 < x < 2 and zero outside that interval. With a = 3 length units per time unit, the pulse occupies 6 < x < 8 at t = 2. Its width and height remain unchanged in this ideal equation. The two sharp edges travel with the same speed.
At the edges, the ordinary derivative is not defined in the classical sense. Nevertheless, the translated pulse satisfies the conservation law in an integral or weak sense. Discontinuity does not automatically mean that the model has failed; it means that the solution concept must match the object being described.
Now compare two observations. If a numerical pulse is lower and wider, that can indicate numerical diffusion. If it has oscillatory undershoots near an edge, that can indicate a different discretisation error. Neither pattern is part of the exact constant-speed transport solution. An exact translation benchmark makes those errors visible without requiring a complex physical experiment.
CHAPTER 3 OF 14 · Follow information and boundaries
3. Supply boundary information only where it enters
Consider u_t + a u_x = 0 on 0 < x < L with a > 0. Characteristics enter through the left boundary and leave through the right. Initial data determine the early solution, while a left-boundary history supplies information that enters later. An arbitrary additional value imposed at the outflow boundary can conflict with the transported interior solution.
Trace a point (x,t) backwards along its characteristic. If x − at ≥ 0, the path reaches the initial line inside the domain, so use the initial value g(x − at). If x − at < 0, it reaches the left boundary at time t − x/a, so use the prescribed inflow history at that time.
A boundary-arrival calculation
Let L = 10, a = 2, initial data be zero, and the left boundary switch from zero to one at t = 1. The leading edge follows x = 2(t − 1). It reaches x = 6 at t = 4. A sensor at x = 6 should therefore remain at zero before that arrival in this ideal model.
If the sign of a changes, the inflow side changes. For a system with several characteristic speeds, some components may enter at each boundary. Counting incoming characteristics is therefore more useful than applying the same number of conditions at every boundary because the domain looks geometrically symmetric.
Corner compatibility also matters. If the initial value at an inflow endpoint differs from the boundary value at time zero, a discontinuity may propagate from that corner. Do not describe the resulting solution as smooth everywhere without checking the data. When a solver produces a new front, first ask whether the initial and boundary specification deliberately created it.
CHAPTER 4 OF 14 · Calculate waves and dependence
4. Derive the string wave speed from a force balance
For a taut uniform string with constant tension T and mass per unit length μ, assume small transverse slopes. The difference in the transverse components of tension across a short segment is approximately T u_xx times the segment length. Newton’s law then gives μu_tt = T u_xx, so c = √(T/μ).
This calculation assumes an ideal flexible string, small displacement gradients, constant tension and no damping or external forcing. Bending stiffness, large stretching or appreciable dissipation can require additional terms or a different model. The simplicity of c = √(T/μ) comes from those assumptions.
Compute a travel time
Let T = 100 newtons and μ = 0.01 kilograms per metre. Their ratio is 10,000 square metres per square second, giving c = 100 metres per second. A disturbance crosses a two-metre string in 0.02 seconds. A round trip between the endpoints takes 0.04 seconds.
If tension is multiplied by four while μ stays fixed, the wave speed doubles and the travel time halves. If both tension and linear density double, the ratio T/μ stays unchanged and so does the ideal wave speed. These parameter comparisons are useful checks before trusting a detailed calculation.
The string’s particles need not travel two metres in 0.02 seconds. They move transversely while the disturbance moves along the string. A marked point can oscillate near its original longitudinal location while a visible crest travels past it. Separating material motion from signal propagation prevents a common error in interpreting waves, sound and related systems.
CHAPTER 5 OF 14 · Calculate waves and dependence
5. Use d’Alembert’s formula with both initial profiles
On the whole line, let u(x,0) = g(x) and u_t(x,0) = h(x), with constant c > 0. The one-dimensional wave solution is the sum of half of each translated displacement profile and an integral of the initial velocity:
u(x,t) = [g(x − ct) + g(x + ct)]/2 + [1/(2c)]∫ from x − ct to x + ct of h(s) ds.
The formula can be checked by differentiating and evaluating at t = 0. The two translated g terms add to g initially, while their initial time derivatives cancel. Differentiating the integral with moving endpoints supplies h. Omitting the integral therefore silently assumes zero initial velocity.
Two initial states that look identical in a photograph
Suppose the initial displacement is zero. In the first experiment, initial velocity is also zero, so the solution remains zero. In the second, let h(x) = 2 on −1 < x < 1 and zero elsewhere, with c = 1. At x = 0 and t = 0.25, the integration interval is [−0.25,0.25], giving u = (1/2)(2 × 0.5) = 0.5.
At x = 0 and t = 2, the interval [−2,2] includes the whole initial velocity pulse, giving u = (1/2)(2 × 2) = 2. These piecewise data are understood in a suitable weak sense at their jumps. The initial photograph of displacement was identical, but the unobserved initial velocity changed the evolution.
Hunter’s chapter on hyperbolic equations develops the wave equation, its initial data and energy framework. Its discussion also contrasts the lack of automatic smoothing with diffusion. The explicit calculations here make that difference testable rather than merely descriptive.
CHAPTER 6 OF 14 · Calculate waves and dependence
6. Draw the domain of dependence before calculating
In the whole-line wave formula, the value at (x,t) uses initial displacement at x − ct and x + ct and initial velocity between those points. The interval [x − ct,x + ct] is the relevant initial domain of dependence. Data outside it cannot affect that particular value within this model.
For c = 2, a point at x = 5 and t = 1.5 depends on initial data from x = 2 to x = 8. Changing the initial velocity near x = 10 does not alter that value. At a later time the interval expands, so the same modification may become relevant.
Distinguish influence from dependence
The domain of dependence asks where the information reaching one point came from. The domain of influence asks which later points can be affected by one initial disturbance. For a disturbance near x = 0, the influence is bounded by the characteristic lines x = ±ct in the ideal constant-speed wave problem.
On a bounded domain, reflections can bring boundary information into the picture. The whole-line formula cannot be applied indefinitely while ignoring endpoints that a wave has already reached. Calculate the earliest boundary arrival and any return travel time before treating a finite string as effectively infinite.
This finite-propagation structure differs from the ideal heat equation’s immediate spreading over an unbounded domain. The Parabolic Partial Differential Equations guide explains diffusion’s smoothing and its model limitations. The distinction is about the equations and their assumptions; it is not a universal claim that every physical process must fit one simple prototype.
When assessing a simulation, identify whether a change appears before it could arrive along the model’s characteristics. An unexpectedly early response can reveal a boundary implementation error, incorrect speed or an inappropriate interpretation of what the plotted variable represents.
CHAPTER 7 OF 14 · Connect energy and nonlinear motion
7. Check energy and reflection rather than expecting decay
For a smooth string solution on an interval, define total mechanical energy as E = ½∫[μu_t² + T u_x²] dx. Differentiate in time, substitute μu_tt = T u_xx and integrate by parts. The interior terms cancel, leaving E’ = [T u_t u_x] at the two ends.
With fixed endpoints u = 0 for all time, u_t also vanishes there and the boundary contribution is zero. Energy is conserved in this ideal undamped model. A computation whose amplitude steadily fades may be introducing numerical dissipation, unless the stated model includes damping or boundary energy loss.
A standing-mode check
Let u(x,t) = A sin(πx/L) cos(cπt/L). It has zero endpoint displacement and zero initial velocity. At t = 0 the energy is entirely in the slope term. Since ∫ sin²(πx/L) dx and ∫ cos²(πx/L) dx both equal L/2, the total energy is T A²π²/(4L) at every time.
At a quarter of the oscillation period, the displacement is zero everywhere but the velocity is largest. Zero displacement at one instant does not mean zero energy. A numerical audit that checks only the height of the string misses this transfer between kinetic and potential contributions.
For an ideal transverse displacement pulse, a fixed endpoint produces an inverted reflected pulse so that incident and reflected displacements cancel at the endpoint. An ideal free endpoint imposes zero slope and produces a non-inverted displacement reflection. These statements are specific to the variable and boundary model; pressure or velocity reflections in acoustics require their own conventions.
Adding a damping term changes the energy identity. For μu_tt + bu_t = T u_xx with b ≥ 0 and no boundary work, E’ = −∫b u_t² dx. That derived loss is physical within the model. It should be separated from extra loss introduced by discretisation.
CHAPTER 8 OF 14 · Connect energy and nonlinear motion
8. Recognise hyperbolicity in a first-order system
For a constant-coefficient system U_t + A U_x = 0, a useful one-dimensional hyperbolicity criterion is that A have real eigenvalues and a complete set of eigenvectors. If A = RΛR⁻¹, the transformed variables W = R⁻¹U satisfy W_t + ΛW_x = 0. Each component travels with its own eigenvalue as characteristic speed.
Real speeds alone are not enough to justify this diagonalisation if the eigenvectors are incomplete. For variable-coefficient or multidimensional systems, the appropriate uniformity and directional conditions also matter. A small constant matrix is an introduction to the mechanism, not a full theorem for every system.
Split a two-component example
Take A = [[0,2],[2,0]]. Its eigenvectors (1,1) and (1,−1) have eigenvalues 2 and −2. If U = (p,v), the combinations w₊ = p + v and w₋ = p − v satisfy transport equations with speeds +2 and −2 respectively.
Start with p(x,0) = g(x) and v(x,0) = 0. Then both characteristic combinations initially equal g. Later, p = [g(x − 2t) + g(x + 2t)]/2 and v = [g(x − 2t) − g(x + 2t)]/2. The initial disturbance separates into two oppositely travelling components.
At a left boundary, the +2 component enters the domain while the −2 component leaves it. At the right boundary, their roles reverse. Prescribing both p and v independently at both boundaries may impose too much information. Formulating boundary data in characteristic variables reveals which information the interior already determines.
For nonlinear conservation systems U_t + F(U)_x = 0, the flux Jacobian replaces A in local characteristic analysis. Its eigenvalues can depend on the state. This dependence is the doorway to wave speeds that change as the solution changes, and to the steepening examined next.
CHAPTER 9 OF 14 · Connect energy and nonlinear motion
9. Calculate nonlinear steepening before a shock forms
Inviscid Burgers’ equation is u_t + (u²/2)_x = 0. While u is smooth, the chain rule gives u_t + u u_x = 0. Its characteristic speed is the value u itself. Along a characteristic starting at ξ, u remains g(ξ), while x = ξ + g(ξ)t.
Different starting values therefore travel at different speeds. Differentiating this mapping with respect to ξ gives ∂x/∂ξ = 1 + g'(ξ)t. When this reaches zero, the smooth characteristic description can cease to be one-to-one. The associated gradient u_x = g'(ξ)/(1 + g'(ξ)t) can become unbounded.
An exact steepening calculation
Take smooth periodic initial data g(ξ) = 1 − 0.2 sin ξ. The most negative slope is −0.2. The first candidate breaking time is therefore −1/min g’ = 5. At the characteristic starting at ξ = 0, the gradient is −0.2/(1 − 0.2t). It equals −0.4 at t = 2.5 and −2 at t = 4.5.
The field values remain finite while the slope grows without bound as t approaches 5 from below. A solver does not necessarily need an infinite value of u to encounter a shock-forming problem. Watching only the range of u can miss the relevant failure of smoothness.
After characteristics intersect, continuing the formula as a multivalued graph does not produce an acceptable single-valued field. A weak conservation-law solution replaces the smooth description, and an admissibility condition is needed to select the appropriate continuation. The following chapters separate those two requirements.
This example is a simplified nonlinear model. It does not mean every hyperbolic PDE forms shocks, nor that every observed sharp front is described by Burgers’ equation. Constant-speed linear transport preserves a smooth profile without this state-dependent steepening mechanism.
CHAPTER 10 OF 14 · Resolve shocks and computation
10. Determine a shock speed from conservation
For u_t + f(u)_x = 0, integrating over a fixed interval gives the rate of change of its total as the incoming flux minus the outgoing flux. A moving discontinuity must respect the same balance. If its left and right states are u_L and u_R, its speed s satisfies s(u_R − u_L) = f(u_R) − f(u_L).
This Rankine–Hugoniot relation is an accounting condition. It determines the speed of a candidate jump, but it does not by itself establish that every such jump is the physically relevant or mathematically admissible solution.
Work through a Burgers shock
Let u_L = 3 and u_R = 1, with f(u) = u²/2. The two fluxes are 4.5 and 0.5. The shock speed is (0.5 − 4.5)/(1 − 3) = 2. A discontinuity initially at x = 0 therefore lies at x = 4 when t = 2.
The characteristic speed on the left is 3 and on the right is 1. The shock travels between them: information approaches the front from both sides. The faster state behind catches the slower state ahead, consistent with compression.
Check the sign by a moving-front balance. In a frame travelling at speed s, the relative flux is f(u) − su. On the left it is 4.5 − 6 = −1.5; on the right it is 0.5 − 2 = −1.5. Equality confirms that the front does not create or destroy the conserved quantity.
The Clawpack Riemann-problem chapter on Burgers’ equation develops shocks, rarefactions and entropy selection. The term Riemann problem refers here to a conservation law with two constant initial states separated by one interface. It is valuable because it isolates the local behaviour used by many numerical flux methods.
CHAPTER 11 OF 14 · Resolve shocks and computation
11. Reject an expansion shock and build a rarefaction
Reverse the Burgers states to u_L = 1 and u_R = 3. The conservation jump formula still gives speed 2. But now the left characteristics travel at 1 and the right characteristics at 3: they separate rather than converge. The admissible scalar convex-flux solution is a rarefaction fan, not the same travelling jump with reversed labels.
For this example, u = 1 where x/t ≤ 1, u = x/t where 1 < x/t < 3, and u = 3 where x/t ≥ 3, for t > 0. Inside the fan, direct differentiation gives u_t = −x/t² and u u_x = (x/t)(1/t), whose sum is zero.
Check three positions at one time
At t = 2, the fan extends from x = 2 to x = 6. At x = 1 the state is 1. At x = 4 the state is 2. At x = 7 the state is 3. A jump moving at speed 2 would instead place a discontinuity at x = 4 and fail to reproduce this spreading family of characteristics.
For scalar Burgers’ equation, the entropy condition distinguishes compressive shocks from expansion jumps. One interpretation comes from adding small positive viscosity and considering an appropriate zero-viscosity limit. This is a selection principle for weak solutions, not permission to add arbitrary smoothing until a numerical plot looks pleasant.
The lesson transfers beyond this one flux: conservation is necessary but may not be sufficient for uniqueness after discontinuities form. The precise entropy framework for systems and more general equations requires further care. Avoid treating the scalar left-speed/right-speed test as a complete universal theory for every hyperbolic system.
When diagnosing a computation, compare both the shock location and the type of wave. A method that conserves the total while replacing a rarefaction with an inadmissible jump can still be solving the wrong weak continuation.
CHAPTER 12 OF 14 · Resolve shocks and computation
12. Make a finite-volume update and inspect the CFL ratio
For cell averages U_i and numerical interface fluxes F_(i+1/2), a conservative update is U_i^(n+1) = U_i^n − (Δt/Δx)[F_(i+1/2) − F_(i−1/2)]. The same interface flux must be used by both adjacent cells. Then internal contributions cancel when the cell balances are added.
For positive constant advection speed a, the first-order upwind flux is a times the value in the cell to the left. The update becomes U_i^(n+1) = (1 − ν)U_i^n + νU_(i−1)^n, with ν = aΔt/Δx. When 0 ≤ ν ≤ 1, these weights are nonnegative and sum to one.
A complete periodic three-cell calculation
Let the old values be [1,0,0], use periodic boundaries and set ν = 0.4. The new values are [0.6,0.4,0]. Their sum remains 1. The pulse is shared between two cells rather than translating sharply by a fractional cell, illustrating the numerical diffusion of this basic scheme.
With ν = 1.2, the new values become [−0.2,1.2,0]. Their sum is still 1, but positivity and the original range have been lost. Conservation alone did not guarantee a useful solution. The negative weight exposed the problem before any lengthy simulation was needed.
The Courant–Friedrichs–Lewy idea relates numerical information flow to the PDE’s domain of dependence. For this explicit upwind scheme the useful condition is the stated ν range. Other stencils, time integrators, dimensions and nonlinear systems have different detailed restrictions. A CFL condition is not a universal certificate of accuracy or sufficient stability for arbitrary schemes.
The existing Finite Volume Method article provides a specialist route into shared fluxes and discrete balance. Use it to connect the control-volume viewpoint with implementation, while retaining separate checks for conservation, wave speed, admissibility and resolution.
Refine the measurement as well as the grid
Suppose a Burgers shock should be at x = 4 at t = 2. A numerical transition spread over several cells needs a declared location estimator. The midpoint crossing, a reconstructed jump position and a mass-based estimate need not agree on a coarse mesh. Record the estimator before comparing methods, then check whether its result approaches the exact position under refinement.
Do not demand pointwise convergence exactly at a discontinuity without specifying what value is intended there. Integral error measures, shock-location error and errors away from the front answer more useful questions. A plot sampled on a different set of points can appear improved even when the underlying approximation has barely changed.
The existing Adaptive Mesh Refinement article provides a route into allocating resolution where the solution needs it. Refinement must still respect the numerical time-step constraints and conservation across changes in resolution. More cells near a shock cannot compensate for an incorrect flux law or entropy treatment.
A practical comparison uses the same initial data, final time, boundary rules and output definitions on every mesh. Report the total conserved amount, the front position and an integral error against an exact Riemann solution. If one improves while another worsens, investigate that tradeoff rather than collapsing all three into a single claim that the simulation is more accurate.
CHAPTER 13 OF 14 · Transfer and continue
13. Transfer the reasoning when the flux changes
Consider the dimensionless traffic-style conservation law ρ_t + f(ρ)_x = 0 with f(ρ) = ρ(1 − ρ), for 0 ≤ ρ ≤ 1. This is a simplified continuum model. Its characteristic speed is f'(ρ) = 1 − 2ρ, which differs from the vehicle-speed expression f(ρ)/ρ = 1 − ρ when ρ > 0.
The distinction is decisive: a pattern can move upstream even while the represented vehicles move downstream. A density front is not a material vehicle. The model’s interpretation depends on aggregation assumptions and does not supply a detailed forecast of individual driver behaviour.
Changed-context exercise
Take left density 0.2 and right density 0.8. Calculate the two fluxes, the candidate shock speed and the two characteristic speeds. Decide whether the front moves, and whether information converges towards it.
Both fluxes equal 0.16, so the jump speed is zero. The left characteristic speed is 0.6 and the right characteristic speed is −0.6. They converge towards the stationary front. The vehicle-speed expressions are 0.8 and 0.2, both positive. A stationary density discontinuity is therefore compatible with positive downstream vehicle motion on both sides.
Now reverse the densities. The same jump formula again gives zero, but the characteristics diverge. Because this flux is concave rather than convex, the orientation of compressive and spreading states differs from the Burgers example. The admissible rarefaction has ρ = (1 − x/t)/2 between x/t = −0.6 and x/t = 0.6, matching 0.8 on the left and 0.2 on the right.
This is a genuine transfer test. Memorising “larger left state means shock” from Burgers fails here. The dependable procedure is to identify the flux, calculate its derivative, check the characteristic directions and apply the admissibility framework appropriate to that flux. Clawpack’s traffic-flow chapter develops this scalar model and its wave interpretation.
CHAPTER 14 OF 14 · Transfer and continue
14. Use a diagnostic and learning route that survives new examples
What should be checked before a numerical run?
State the field and units, equation, domain, initial data, incoming boundary data and output of interest. Calculate the characteristic speeds and a relevant travel time. Decide whether smooth solutions are expected throughout the interval or whether a shock may form. Finally choose an exact benchmark and a conservation or energy check that follows from the actual model.
For transport, a translated pulse is an excellent first benchmark. For the wave equation, a standing sine mode checks phase and energy. For nonlinear conservation, a two-state Riemann problem checks shock speed or rarefaction structure. These tests answer different questions and should not be replaced by one visually complicated demonstration.
Does hyperbolic mean every solution forms a shock?
No. Linear constant-speed transport simply translates its initial field. The ideal linear wave equation superposes travelling components. Nonlinear characteristic speeds can create steepening, but the outcome depends on the equation and data. Smoothness, discontinuity and physical breakdown should be discussed separately.
Why is my wave arriving at the wrong time?
Check units and the physical speed first, then inspect boundary reflections and numerical phase error. A method can preserve amplitude reasonably well while transporting a crest at the wrong numerical speed. Compare arrival times at multiple resolutions rather than judging only the final shape.
Why can a conservative calculation still be wrong?
The three-cell example conserves its sum even with a negative value. A conservative scheme can also have excessive diffusion, dispersive oscillations or an incorrect entropy treatment. Conservation is one necessary piece of evidence; it does not replace positivity, admissibility, convergence and model checks.
How should a learner practise?
First trace characteristics for positive and negative transport velocities. Then use both initial profiles in the wave formula. Next calculate a Burgers shock and reverse the states to build the rarefaction. Finally change to the concave traffic flux and explain why the memorised state-order rule changes. Write the expected direction and speed before computing numbers.
For further study, Hunter’s hyperbolic-equation notes provide the wave and energy framework. The Clawpack Riemann book connects characteristics, conservation, shocks and numerical solvers. Return to the Partial Differential Equations guide for the broader classification or the Gold Standard master for another subject route.