A partial differential equation relates derivatives in several independent variables. Its solutions form a family; initial or boundary data select a member. Before applying a formula, we must ask where the data are prescribed and whether they actually determine a solution nearby.
Transport along characteristics
Consider the transport equation on with , where is constant. Along a curve , the chain rule gives
Choosing makes this derivative zero, so a solution is constant along every line .
Theorem 1.1.1
(Solution of the transport equation)
Let and . The problem for , , has exactly one solution , namely
Proof
(Constant along lines, then verify)
Uniqueness: let be a solution and fix . The function on has derivative , so it is constant. Its value at is and at is . Existence: for the chain rule gives and , so , and .
A profile travels without changing shape.
Worked example 1.1.2
(A traveling Gaussian)
Take and . Then . At time the maximum value sits at , and, for instance, because the point lies on the characteristic through . When every characteristic is vertical and the profile stays where it is: the edge case .
The direction matters. For on , , characteristics enter through . Initial data alone do not determine values at points with ; boundary values are needed there. When , characteristics leave at , and arbitrary boundary data can overdetermine the problem.
Common mistake
(Prescribing data where characteristics leave)
A boundary condition at is part of the problem only where characteristics enter the domain. For every characteristic reaching already carries a value from the initial line, namely . Prescribing a different there makes the problem unsolvable rather than better determined.
General first-order equations
For , with coefficients independent of , the same chain-rule idea follows curves along which the left side is a total derivative.
Definition 1.1.3
(Characteristics and noncharacteristic data)
The characteristic system of is
where records the value of along the projected curve . Data on a curve are noncharacteristic at if
which is the Jacobian of at .
Concept lensGeometric
Characteristics form a local coordinate grid
The parameter labels the starting characteristic; moves along that characteristic. Reconstructing means using these two labels as coordinates for the physical point. The nonzero Jacobian says their directions remain independent at the initial curve. Local inversion then assigns one evolved solution value to each nearby point. Without it, the same line of data can either demand inconsistent values or leave transverse values undetermined.
A nonzero value of means the characteristic direction is transverse to the initial curve. That is exactly what the local theorem needs.
Theorem 1.1.4
(Local solvability for noncharacteristic data)
Let , the initial curve and be near , and let . Then there is a neighborhood of on which has a solution with for near , and any two solutions with these data agree near that point.
Proof
(Solve the ordinary equations, then invert)
Solve the characteristic system with initial values at . The ODE existence theorem with dependence on initial data gives a map for small and near . Its part has Jacobian at , so the inverse function theorem gives a inverse near the point. Define . On each characteristic, the chain rule gives
while the left side is ; hence . At the data hold by construction. For uniqueness, restrict any solution to a characteristic segment through the neighborhood: it satisfies with the same initial value, so it equals there.
Proof X-ray
Separate the local and global questions
The chain-rule identity is the one move that turns a PDE into ODEs: along a characteristic, is just . The hypotheses are used only to make the characteristic flow a function of . Transversality is used only once, in the inverse function theorem, and it is what gives each nearby point a single label . That inverse is local around one starting point. It does not make the parametrization globally one-to-one: two different labels can lie on the same characteristic, and since the PDE transports one value along it, data at both labels must agree if a solution is to exist along the connecting segment. Distinct projected trajectories of this autonomous linear equation cannot cross by ODE uniqueness. For quasilinear equations whose characteristic speeds depend on , crossing is a separate possible phenomenon.
The construction is easiest to follow on an equation with a source.
Worked example 1.1.5
(A constant source term)
Solve with . Here and the data curve is , so everywhere. Characteristics are and . Since and , the solution is . Direct substitution gives and confirms the data. Here the inversion is global, because every characteristic meets the line exactly once.
When along the whole data curve, the theorem gives nothing, and two different failures can occur.
Counterexample mode
Characteristic data can fail in two different ways
Transversality is not a decorative condition. For with data on , the initial curve is itself characteristic: . The data are incompatible because a solution is constant along that line, yet the data take the two values and at and . The data admit infinitely many solutions with and , for example and .
A curved data curve: local branches and a fold
The transversality test becomes useful when the starting curve bends.
Worked example 1.1.6
(Data on a parabola)
Consider
We first seek a solution near one chosen point of the parabola. The characteristic starting at is , and its carried value is . The coordinate map and its Jacobian, with columns ordered as , are
Thus it can be inverted locally whenever . To perform the inversion, subtract the two coordinates:
If the chosen initial point has , the minus sign returns that point's label; for , the plus sign does. On the open half-plane , the two resulting functions
are smooth and each satisfies the PDE because it depends only on . On the corresponding half of the parabola, its chosen branch satisfies the data. On a sufficiently small neighborhood of any one noncharacteristic point, every solution with those local data agrees with the appropriate branch, by the uniqueness part of the theorem.
The two branches cannot be combined into a single solution satisfying the data on the whole parabola. For example, the distinct points and lie on the same characteristic , but the data demand the respective values and . Their conflict concerns a domain containing the connecting characteristic segment; local existence near either point is unaffected. The fold at has an even sharper obstruction. If a solution existed near and obeyed the data near , differentiating gives , which at reads
contradicting the PDE there. This edge case shows the theorem's hypothesis failing at a single point of an otherwise transverse curve.
Common mistake
(Reading a zero Jacobian as nonexistence)
A vanishing Jacobian does not by itself imply that no solution exists: it tells us only that the coordinate argument cannot determine a local solution there. In the parabola example, nonexistence near came from the separate incompatibility forced by these particular data. Other data on the same curve can be perfectly solvable, as the next check shows.
Quick check
(Keep and the parabola , but prescribe . Find a smooth solution on the plane and explain why the zero Jacobian at does not obstruct it.)
Look for a function of .
Solution · Answer
Take . Then , and on the parabola . The data agree at labels sharing a characteristic, and at the tangent point their derivative is zero, as the equation requires. The singularity of this particular coordinate map prevents its direct inversion there; it does not say that every compatible solution is singular.
What well-posedness says
Definition 1.1.7
(Well-posed problem)
A problem is well posed in specified function spaces if a solution exists, is unique, and depends continuously on the data in the specified norms.
Each part is necessary, and each can fail independently: the characteristic-data counterexample above shows failure of existence and failure of uniqueness for the same equation.
Proposition 1.1.8
(Sup-norm stability of transport)
If and solve on with bounded initial data and , then for every
Proof
(Translation is a bijection)
By the transport theorem, . As ranges over , so does , so the two suprema are taken over the same set of values.
Thus, within bounded initial data and the sup norm, perturbations do not grow. A different equation, domain, solution class, or norm can change the conclusion; “well posed” without these specifications is incomplete.
Practice with solutions
Quick check
(Solve , .)
Shift the profile along .
Solution · Answer
. Check: and , so .
Quick check
(Solve with .)
Find the quantity that is constant along , .
Solution · Answer
The characteristics have constant, so and the data give : . Substitution gives .
Quick check
(For on , with and , determine the solution separately for and .)
Follow the characteristic through backward until it leaves the quadrant.
Solution · Answer
The backward curve is . For it reaches , so . For it meets , so . Continuity on requires the corner compatibility ; a classical solution can require further compatibility of derivatives.