A transport equation moves information along characteristic paths. On the whole line, tracing a path backwards always reaches the initial line. On a finite interval it can reach an endpoint first. The data needed at that endpoint depend on the direction of travel; prescribing values independently at both endpoints can demand two different values on the same path.
The first data point seen backwards
Let , , and consider the real problem
with initial data for and inflow data for . We assume , and .
Definition 1.3.1
(Inflow boundary and classical solution)
For (1) with , the endpoint is the inflow boundary, where characteristics enter the rectangle, and is the outflow boundary, where they leave. A classical solution of (1) with data is a function such that are continuous on , equation (1) holds for and , for , and for .
The characteristic arriving at is . Moving backwards, it hits the initial line at when ; if , it instead hits the left endpoint at time . The point where these alternatives meet, , is the characteristic issued from the lower-left corner. It exists inside the rectangle only where .
Along the path, obeys . Multiplying by and integrating gives the only possible solution:
The two formulas agree on the common interface exactly when their data agree at the corner. No data are prescribed at : since , characteristics leave there. The solution determines the outflow trace, rather than choosing it independently.
Corner compatibility and a classical solution
Continuity of both traces at requires . If first derivatives also extend continuously to that corner, the PDE requires
The value condition aligns values; the derivative condition (3) aligns derivatives. They are also sufficient for the regularity just specified.
Theorem 1.3.2
(Classical inflow problem)
Under the stated assumptions and both corner conditions and (3), formula (2) defines the unique classical solution on the closed rectangle. If only the value condition holds, (2) still gives a continuous function which is separately on the two open sides of , and it is across that line if and only if (3) holds.
Proof
(Matching the two characteristic formulas)
Away from , the formulas are by differentiation under the integral, including the moving lower endpoint. Differentiation or the characteristic ODE verifies (1), and the formulas recover at and at . When , the two integrals agree and the remaining terms agree exactly when .
To compare derivatives, write and evaluate both one-sided derivatives at . Their integral contributions from differentiating the integrand agree. The initial-side contribution is ; the boundary-side extra terms are
Condition (3), with , makes these equal. Each side satisfies , so the one-sided derivatives also agree. These matching continuous derivatives give a function across the interface, with one-sided derivatives on the rectangle boundary. Uniqueness follows by integrating the scalar ODE along the backward characteristic of any other classical solution. With only the value condition the piecewise formula remains continuous. At an interior interface point its boundary-side minus initial-side derivative equals ; the corresponding -derivative jump is times this value. Thus (3) is necessary as well as sufficient for matching.
The proof does three separate jobs, and only one of them uses the corner conditions.
Proof X-ray
Why is there a derivative condition?
Existence away from the interface is routine: each branch of (2) solves the scalar ODE along a characteristic, and differentiating under the integral needs only . The one non-routine move is the comparison on . Differentiating the boundary branch in produces, besides the common integral, three extra terms: from , from the factor , and from the moving lower limit . All three are evaluated at the corner, so they collect into ; this is the only place where (3) is used. The derivatives need no separate computation, because the PDE expresses through , and , which already match. Uniqueness uses neither corner condition: any classical solution restricted to a characteristic solves the same linear ODE with the same starting value. Without (3), the jump in equals a fixed constant times , so the corner mismatch is carried along the whole interface and is never smoothed out.
A piecewise solution and an outflow trace
Worked example 1.3.3
(Matched quadratic data and the outflow trace)
Take , , , and . Both corner conditions hold. Formula (2) becomes
Both the value and first derivatives vanish on ; the second derivatives do not match there: equals 2 on the initial side and on the boundary side. At the outflow endpoint, the data give the trace
This is an output of the problem. An additional right-end condition can be imposed only if it agrees with this trace; it is not a free second boundary input.
A source is accumulated along the same paths, whichever side of the interface they start from.
Worked example 1.3.4
(A constant source on both kinds of path)
Let , , and . The corner conditions hold: , and . For , formula (2) gives . For , the path starts on the boundary at time with value , and the source adds the travel time . Both branches give , which indeed satisfies , and .
Stability, sign and boundary energy
Formula (2) is linear in the data, which makes stability and sign questions direct.
Proposition 1.3.5
(Sup-norm stability and preservation of sign)
Let and be the classical solutions for compatible data and , and write , , , . If , then
For every real , if , then .
Proof
(Bounding each term of the characteristic formula)
By linearity, is given by (2) with data . Since , the exponential factors in (2) are at most 1, and each integration interval has length at most . Hence the data term is bounded by on one side of the interface and by on the other, and the integral by ; this is (6). For the sign, each term of (2) is a strictly positive exponential times a value of , or , or an integral of such products; the exponential is positive for every real .
This is a continuous-dependence bound in the sup norm, in the classical class and for compatible data. The assumption is used only for (6). An arbitrary signed source does not preserve positivity.
The same direction appears in an balance. For a classical solution, multiply (1) by and integrate in :
Incoming data contribute at the left and outgoing energy leaves at the right. If , the energy cannot increase; a nonzero incoming signal can raise it, so a blanket decay claim for arbitrary boundary data would be wrong.
What fails without compatibility?
If , no function continuous at the corner can have both traces. Even when the values agree, first derivatives may fail.
Worked example 1.3.6
(Values match but derivatives do not)
For , take and . Then , and (2) gives
This function is continuous, but its derivative jumps from 0 to 1 across . Condition (3) fails because . The jump agrees with the general formula from the proof: for boundary side minus initial side. It is not a classical solution, although each side separately satisfies the equation. A weak-solution theory for such data requires its own definition and is not established here.
A second error goes the other way: it produces a smooth function by ignoring the boundary data.
Common mistake
(Using the initial-data branch beyond the corner characteristic)
For the expression evaluates at a negative argument, where no data are given. Extending by its formula gives a function that solves the PDE but ignores . With , , , the extension has . Points with must use the boundary branch of (2).
The opposite temptation is to prescribe data at both ends.
Counterexample mode
Two endpoint values are not two independent inputs
For and , the corner conditions hold. The right trace is zero for and afterwards. Demanding an identically zero right trace at all times contradicts the characteristic formula. This argument applies whenever the time interval actually contains such times.
Practice with solutions
Quick check
(Let , , and . Compute and verify the corner conditions.)
Solution · Answer
For , ; for , . Both branches give . The values agree at 1, and .
Quick check
(For , , , , , calculate on both sides of the interface.)
Solution · Answer
Both branches give . The initial branch integrates for time ; the boundary branch adds the starting value to the travel time .
Quick check
(If , , and , do the value and derivative corner conditions hold? Is the formula ?)
Solution · Answer
The values agree at 0, but . The formula is and is not on inside the rectangle.
Quick check
(Two compatible solutions have the same source, and . Bound their difference uniformly when .)
Solution · Answer
In (6) the source difference is zero, so .