Initial values become algebraic terms
The Laplace transform replaces differentiation in time by multiplication in a new variable, but it also records the starting values. This is especially helpful when a forcing switches on after the initial time: a delayed input becomes an exponential factor. We must still check that the integrals converge and that the reconstructed function has the regularity required by the original equation.
Definition 1.13.1
(Exponential order and the Laplace transform)
A function on is piecewise continuous if each bounded interval contains only finitely many jumps and has finite one-sided limits at each of them. It has exponential order if for and some constant . For such and real , the Laplace transform is
The integral converges absolutely, bounded by . A function may have a transform on a larger set than this estimate gives; a growth bound is sufficient, not a statement of the exact convergence boundary. We work with real here, so no complex inversion formula is needed. Two basic transforms used throughout are and for .
Differentiation and the starting values
Theorem 1.13.2
(Transform of a derivative)
Suppose and are continuous and piecewise continuously differentiable on , with finitely many breakpoints on bounded intervals, and that have exponential order, with taken on the smooth pieces. Then, for every real larger than the growth exponents,
Proof
(Integration by parts on the smooth pieces)
Fix and let contain the breakpoints of in . On each piece is continuously differentiable, so
Because is continuous, the value at each interior enters once with a plus sign and once with a minus sign, so the sum telescopes to . If and , then . Letting gives the first formula. The function is again continuous and piecewise continuously differentiable, so the first formula applies to it: .
Proof X-ray
Where continuity and growth enter
The initial-value terms come from the left endpoint of the integration by parts, before any equation is solved. Continuity is used only to cancel the boundary terms at interior breakpoints. If jumped at some , the second application would leave the uncancelled term , so (2) would acquire an extra summand. The growth bound is used only at the right end, where the exponential weight must dominate the function whose derivative is transformed; without it, need not tend to zero.
Common mistake
(Dropping the initial values)
Writing is correct only when . For one has , while ; the difference is exactly . In an initial-value problem these terms carry the given data into the algebraic equation.
Convolution and a delayed input
Definition 1.13.3
(Convolution and the delayed step)
For piecewise continuous functions of exponential order, the convolution on the half-line is
For , the delayed unit step is for and for .
The value of at the jump does not affect any transform, because a single point does not change an integral.
Theorem 1.13.4
(Convolution and delay rules)
If exceeds the growth exponents of and , then
Proof
(Fubini on a quarter plane and a shift of the time origin)
Let and . For the double integral of over is at most , so Fubini's theorem allows any order of integration. The change of variables maps the quarter plane onto , and
which is (3). Isolated jumps do not affect the integrals. For (4), only contributes, and gives .
For example, for . The factor is a delay in time, not a translation ; the latter corresponds to multiplying by .
Common mistake
(Delaying the step but not the function)
Rule (4) requires the function after the switch to be written in the variable . For instance , so
not , which is the transform of .
Why the transform determines a piecewise continuous function
Theorem 1.13.5
(The transform determines the function at continuity points)
Suppose piecewise continuous and have exponential order and for every sufficiently large real . Then at every at which both are continuous. In particular, continuous functions in this class with equal transforms are identical.
Values assigned at isolated jumps are invisible to the integral, so nothing better can be expected at a jump.
Proof
(Moments on the unit interval and polynomial approximation)
Set , which is piecewise continuous with . Choose above the threshold where the transforms agree, and let
Since , the function is bounded on , and it is continuous wherever is continuous at . Substituting , , gives
By the Weierstrass approximation theorem every continuous on is a uniform limit of polynomials, and is bounded, so . If at a continuity point , then keeps one sign on a small interval around ; a nonnegative continuous bump supported there, positive at , makes , a contradiction. Hence , and therefore , vanishes at every continuity point.
Proof X-ray
Turning a transform into moments
The proof has one key move: the substitution converts the values of at the equally spaced points into the moments of a single function on a compact interval. The exponential-order bound is used once, in choosing ; it is what makes bounded near , which corresponds to . Compactness of is what lets polynomials approximate every continuous test function uniformly. The argument closes by localization: a vanishing integral against every bump forces wherever is continuous, and nothing is concluded at jumps.
Worked examples
Worked example 1.13.6
(Switching on a constant force)
Consider with and . Seek a solution that is continuously differentiable on and twice continuously differentiable away from , satisfying the equation at every . The forcing jumps, so the equation is not required to hold classically at .
For such a solution of exponential order, (2) and (4) give
By the basic transforms and (4), the candidate is
Verify rather than relying solely on algebra: before , both and are zero; after , and , so . At , the function and first derivative agree from both sides and equal zero. The second derivative jumps from zero to one, so this solution is not at the switching point. Uniqueness for linear initial-value problems on and on gives uniqueness in the stated class, including among solutions not assumed in advance to have exponential order: on each side the difference solves the homogeneous equation with zero initial data.
The jump in is not a defect of the method but a necessity, as the following shows.
Counterexample mode
A step forcing cannot have a globally twice-continuous solution
Claim: a solution produced by the transform is automatically . If and , then is continuous at . It cannot equal the step's left value zero for all and its right value one for all . The transform result (5) is a legitimate piecewise classical solution, not a globally one.
Worked example 1.13.7
(Nonzero initial data and partial fractions)
Solve with , . By (2), with and ,
The coefficients come from evaluating divided by the other two factors at : , and . Hence
Check: and . In the coefficient of is , that of is , and that of is . The initial values entered the transformed equation through the term , exactly as (2) prescribes.
Worked example 1.13.8
(A pulse of length leaves the oscillator at rest)
Let be a unit pulse on and solve with zero initial data. Then , and since , rule (3) gives the convolution formula
For this is , which is (5) with delay . For it is , the same as from (4). In the edge case the response after the pulse is : the pulse ends exactly when and , so uniqueness keeps the oscillator at rest for all later time.
Practice
Quick check
(Compute for real and state a sufficient range of .)
Integrate the exponential directly.
Solution · Answer
for ; the integral diverges for .
Quick check
(Derive the transform of in terms of , and under the hypotheses of (2).)
Apply both formulas in (2) and collect terms.
Solution · Answer
.
Quick check
(For , find the solution of , , which is continuous and differentiable away from . Check its behavior at the switching time.)
Use .
Solution · Answer
. It is zero before and solves after . Both one-sided values at are zero, but the one-sided derivatives are and . It is continuous, not differentiable at the switch.
Quick check
(The constant functions and have convolution . Confirm (3) for them.)
Compute and .
Solution · Answer
. Since and for , both sides of (3) equal .