Transforms of Derivatives and IVPs

Can we solve a differential equation without ever integrating? This question sounds absurd. Then you meet the Laplace transform. It is a device that trades calculus for algebra. The key mechanism is a formula. It converts differentiation in tt into multiplication by ss. This holds up to a correction term built from initial data. This single fact turns an initial value problem into something new. An initial value problem is a differential equation plus initial conditions. It becomes a linear algebraic equation for the transformed unknown function. The idea matters because it unifies two separate tasks. Classical methods handle these tasks separately: finding a general solution, then fitting constants to initial conditions. The Laplace transform bakes the initial conditions directly into the transformed equation. So the constants of integration never appear as free parameters. They are not solved for later. Instead, they enter as explicit numbers the moment we transform y′y' and y′′y''. The big picture is a three-step algorithm. First, transform the ODE and initial conditions into an algebraic equation in ss. Second, solve algebraically for Y(s)=L{y}(s)Y(s) = \mathcal{L}\{y\}(s). Third, invert the transform to recover y(t)y(t). This article builds the middle bridge of that algorithm. It covers the transform rules for derivatives, with complete proofs. It also shows exactly how an IVP becomes algebra. By the end, you will have several things. You will have a rigorous formula for L{y(n)}\mathcal{L}\{y^{(n)}\} in terms of L{y}\mathcal{L}\{y\} and initial data. You will have a full derivation of the algebraic equation for constant-coefficient linear IVPs. You will also have worked examples carried through to a completely solved Y(s)Y(s). The remaining step — recovering y(t)y(t) from Y(s)Y(s) — is the subject of Inverse Laplace Transform and Partial Fractions.

Key takeaways
  • The Laplace transform of a derivative is L{f′}(s)=sF(s)−f(0)\mathcal{L}\{f'\}(s) = sF(s) - f(0), trading a derivative for multiplication by ss.
  • Iterating this gives L{f′′}(s)=s2F(s)−sf(0)−f′(0)\mathcal{L}\{f''\}(s) = s^2F(s) - sf(0) - f'(0), and a general formula for order nn.
  • Hypotheses matter: ff must be continuous (or piecewise continuous with matching jump terms), f′f' piecewise continuous, and both of exponential order so the transform exists.
  • Applying this to a linear IVP with constant coefficients converts it instantly into a single algebraic equation for Y(s)=L{y}(s)Y(s) = \mathcal{L}\{y\}(s).
  • Solving that algebraic equation gives Y(s)Y(s) as a rational function of ss built entirely from the coefficients and initial values.
  • Key idea: differentiation in tt becomes multiplication by ss (plus initial-data correction terms) in the transform domain, turning calculus into algebra.

Where this lives today

Electrical engineers use exactly this derivative-transform rule to convert RLC circuit equations (a second-order linear ODE relating voltage and current) into algebraic impedance equations in the "ss-domain," the backbone of transfer-function analysis and control theory. Mechanical and structural engineers apply the identical algebraization to spring–mass–damper systems, forced vibration problems, and beam-deflection ODEs. In control theory, the transform of derivatives underlies the definition of the transfer function H(s)=Y(s)/U(s)H(s) = Y(s)/U(s) for a linear time-invariant system, letting engineers analyze stability via the locations of poles (roots of as2+bs+cas^2+bs+c) without ever solving the ODE in the time domain. In signal processing, the closely related zz-transform for discrete-time systems mirrors this exact algebraization idea for difference equations. In probability and statistics, the moment generating function and characteristic function play an analogous algebraizing role for random variables, converting convolution of densities into multiplication of transforms — a conceptual cousin of turning differentiation into multiplication by ss. The technique also underlies numerical and symbolic ODE solvers that use transform methods as a preprocessing step before an inverse-transform (partial fraction or contour integral) routine recovers the time-domain solution.

Ask the tutor: How does this algebraization change if the initial conditions are given at a nonzero time t0t_0, or if the equation has variable coefficients like t y′′t\,y''?

DefinitionLaplace transform of a derivative

Let f:[0,∞)→Rf:[0,\infty)\to\mathbb{R} be continuous on [0,∞)[0,\infty), let f′f' be piecewise continuous on every finite interval [0,A][0,A], and suppose there exist constants M>0M>0, K>0K>0, a∈Ra\in\mathbb{R} such that

∣f(t)∣≤Keatfor all t≥M|f(t)| \le K e^{at} \quad \text{for all } t \ge M

(i.e. ff is of exponential order aa). Then the Laplace transform

F(s)=L{f(t)}(s)=∫0∞e−stf(t) dtF(s) = \mathcal{L}\{f(t)\}(s) = \int_0^\infty e^{-st} f(t)\, dt

converges for s>as > a, and the derivative f′f' has Laplace transform

L{f′(t)}(s)=sF(s)−f(0),s>a.\mathcal{L}\{f'(t)\}(s) = sF(s) - f(0), \qquad s > a.

More generally, if f,f′,…,f(n−1)f, f', \dots, f^{(n-1)} are continuous on [0,∞)[0,\infty) and of exponential order aa, and f(n)f^{(n)} is piecewise continuous on every finite interval, then for s>as>a,

L{f(n)(t)}(s)=snF(s)−sn−1f(0)−sn−2f′(0)−⋯−f(n−1)(0).\mathcal{L}\{f^{(n)}(t)\}(s) = s^n F(s) - s^{n-1} f(0) - s^{n-2} f'(0) - \cdots - f^{(n-1)}(0).

Intuition

Think of the Laplace transform as a machine that takes a signal f(t)f(t) living on t≥0t \ge 0 and reports back a "frequency-like" profile F(s)F(s). The transform is built from the exponential kernel e−ste^{-st}, and exponentials are the unique functions whose derivative is a multiple of themselves. That self-similarity is exactly why differentiating ff before transforming is almost the same as multiplying F(s)F(s) by ss after transforming — the kernel absorbs the derivative through integration by parts, leaving behind only a boundary term. That boundary term is the initial value f(0)f(0). This is the heart of why the method is so well suited to initial value problems: the transform of y′y' does not just involve Y(s)Y(s), it involves Y(s)Y(s) and the number y(0)y(0), automatically encoding the initial condition into the algebra. Differentiate twice, and both y(0)y(0) and y′(0)y'(0) appear. No separate step of "solve for constants using initial conditions" is needed — the constants are already sitting inside the equation for Y(s)Y(s). Picture the transform as sliding differentiation from the tt-side of the ledger to the ss-side, at the cost of a fixed toll paid in initial data.

Figure
Differentiation in tt becomes multiplication by ss, minus a toll for initial data
t-domain s-domain f(t), f′(t) F(s), sF(s) − f(0) Laplace transform L differentiate: f → f′ multiply by s, subtract toll f(0) toll paid once per derivative order

Central theorem

TheoremTransform of the nn-th derivative

Let n≥1n \ge 1 be an integer. Suppose: 1. f,f′,f′′,…,f(n−1)f, f', f'', \dots, f^{(n-1)} are continuous on [0,∞)[0,\infty); 2. f(n)f^{(n)} is piecewise continuous on every finite subinterval [0,A][0,A], A>0A>0; 3. there exist constants K>0K > 0, a∈Ra \in \mathbb{R}, M≥0M \ge 0 such that

∣f(t)∣, ∣f′(t)∣, …, ∣f(n−1)(t)∣≤Keatfor all t≥M|f(t)|,\ |f'(t)|,\ \dots,\ |f^{(n-1)}(t)| \le K e^{at} \quad \text{for all } t \ge M

(all lower derivatives are of exponential order aa). Then F(s)=L{f(t)}(s)F(s) = \mathcal{L}\{f(t)\}(s) exists for s>as > a, and for s>as > a,

L{f(n)(t)}(s)=snF(s)−∑k=0n−1s n−1−kf(k)(0).\mathcal{L}\{f^{(n)}(t)\}(s) = s^n F(s) - \sum_{k=0}^{n-1} s^{\,n-1-k} f^{(k)}(0).

Explicitly, for n=1n=1 and n=2n=2 (the cases used for second-order IVPs):

L{f′(t)}(s)=sF(s)−f(0),\mathcal{L}\{f'(t)\}(s) = sF(s) - f(0),
L{f′′(t)}(s)=s2F(s)−sf(0)−f′(0).\mathcal{L}\{f''(t)\}(s) = s^2 F(s) - s f(0) - f'(0).
Proof

Technique: induction on nn, with the base case proved by integration by parts on a truncated interval followed by a limiting argument (to legitimately handle the improper integral and any jump discontinuities of f(n)f^{(n)}). Base case n=1n=1. We must show L{f′}(s)=sF(s)−f(0)\mathcal{L}\{f'\}(s) = sF(s) - f(0) for s>as>a, under hypotheses (1)–(3) with n=1n=1: ff continuous on [0,∞)[0,\infty), f′f' piecewise continuous on every [0,A][0,A], and ff of exponential order aa. Fix s>as > a. Because f′f' is piecewise continuous on [0,A][0,A] for every A>0A>0, it has at most finitely many discontinuities on [0,A][0,A], say at 0<t1<t2<⋯<tm<A0 < t_1 < t_2 < \cdots < t_m < A. Split the integral ∫0Ae−stf′(t) dt\int_0^A e^{-st} f'(t)\,dt at these points and integrate by parts on each subinterval, where f′f' is continuous and ff is continuously differentiable (using one-sided derivatives at endpoints where necessary, which suffices since ff itself is continuous across the tit_i by hypothesis):

∫0Ae−stf′(t) dt=[e−stf(t)]0A+s∫0Ae−stf(t) dt.\int_0^A e^{-st} f'(t)\, dt = \left[e^{-st}f(t)\right]_0^A + s\int_0^A e^{-st} f(t)\, dt.

This identity holds because integration by parts is applied piece by piece on [0,t1],[t1,t2],…,[tm,A][0,t_1],[t_1,t_2],\dots,[t_m,A], and the boundary terms at the interior points t1,…,tmt_1,\dots,t_m cancel telescopically: the "−e−stf(t)-e^{-st}f(t)" evaluated at the right end of one piece equals the negative of the same quantity evaluated at the left end of the next piece (since ff, unlike f′f', is continuous there). What survives is exactly the outer boundary evaluation at t=0t=0 and t=At=A. Hence

∫0Ae−stf′(t) dt=e−sAf(A)−f(0)+s∫0Ae−stf(t) dt.\int_0^A e^{-st} f'(t)\, dt = e^{-sA} f(A) - f(0) + s\int_0^A e^{-st} f(t)\, dt.

Now let A→∞A \to \infty. Since ff is of exponential order aa and s>as > a, choose MM as in hypothesis (3); for A>MA > M,

∣e−sAf(A)∣≤Ke−sAeaA=Ke−(s−a)A→0as A→∞,\left|e^{-sA} f(A)\right| \le K e^{-sA} e^{aA} = K e^{-(s-a)A} \to 0 \quad \text{as } A \to \infty,

because s−a>0s - a > 0. Also, by definition of the Laplace transform's convergence for s>as>a (guaranteed since ff is continuous, hence piecewise continuous, and of exponential order aa — the existence hypothesis inherited from The Laplace Transform: Definition and Existence),

∫0Ae−stf(t) dt  ⟶  F(s)as A→∞.\int_0^A e^{-st} f(t)\, dt \;\longrightarrow\; F(s) \quad \text{as } A \to \infty.

Finally we must confirm the left side converges, i.e. that L{f′}(s)\mathcal{L}\{f'\}(s) exists. But this follows immediately from the identity itself: as A→∞A\to\infty the right-hand side converges (both terms converge, as just shown), so the left-hand side ∫0Ae−stf′(t) dt\int_0^A e^{-st}f'(t)\,dt converges to the same limit; this is precisely the statement that the improper integral defining L{f′}(s)\mathcal{L}\{f'\}(s) converges. Taking the limit A→∞A\to\infty in the identity gives

L{f′}(s)=0−f(0)+sF(s)=sF(s)−f(0),\mathcal{L}\{f'\}(s) = 0 - f(0) + sF(s) = sF(s) - f(0),

proving the base case. Inductive step. Suppose the theorem holds for some n=j≥1n=j\ge 1: whenever g,g′,…,g(j−1)g,g',\dots,g^{(j-1)} are continuous and of exponential order aa and g(j)g^{(j)} is piecewise continuous on finite intervals,

L{g(j)}(s)=sjG(s)−∑k=0j−1sj−1−kg(k)(0),s>a.\mathcal{L}\{g^{(j)}\}(s) = s^j G(s) - \sum_{k=0}^{j-1} s^{j-1-k} g^{(k)}(0), \qquad s>a.

Now let ff satisfy the hypotheses for n=j+1n = j+1: f,…,f(j)f,\dots,f^{(j)} continuous and of exponential order aa, and f(j+1)f^{(j+1)} piecewise continuous on finite intervals. Apply the base case (n=1n=1) to g:=f(j)g := f^{(j)}: since g=f(j)g=f^{(j)} is continuous and of exponential order aa (hypothesis for n=j+1n=j+1 includes this for k=jk=j), and g′=f(j+1)g' = f^{(j+1)} is piecewise continuous, we get

L{f(j+1)}(s)=s L{f(j)}(s)−f(j)(0).\mathcal{L}\{f^{(j+1)}\}(s) = s\,\mathcal{L}\{f^{(j)}\}(s) - f^{(j)}(0).

Apply the inductive hypothesis to g=fg=f itself (with the derivative order jj; note f,…,f(j−1)f,\dots,f^{(j-1)} continuous of exponential order aa and f(j)f^{(j)} piecewise continuous — all inherited from the n=j+1n=j+1 hypotheses) to expand L{f(j)}(s)\mathcal{L}\{f^{(j)}\}(s):

L{f(j)}(s)=sjF(s)−∑k=0j−1sj−1−kf(k)(0).\mathcal{L}\{f^{(j)}\}(s) = s^j F(s) - \sum_{k=0}^{j-1} s^{j-1-k} f^{(k)}(0).

Substitute:

L{f(j+1)}(s)=s(sjF(s)−∑k=0j−1sj−1−kf(k)(0))−f(j)(0)\mathcal{L}\{f^{(j+1)}\}(s) = s\left(s^j F(s) - \sum_{k=0}^{j-1} s^{j-1-k} f^{(k)}(0)\right) - f^{(j)}(0)
=sj+1F(s)−∑k=0j−1sj−kf(k)(0)−s0f(j)(0)=sj+1F(s)−∑k=0js(j+1)−1−kf(k)(0),= s^{j+1}F(s) - \sum_{k=0}^{j-1} s^{j-k} f^{(k)}(0) - s^{0} f^{(j)}(0) = s^{j+1}F(s) - \sum_{k=0}^{j} s^{(j+1)-1-k} f^{(k)}(0),

which is exactly the claimed formula for n=j+1n = j+1. By the principle of mathematical induction, the formula holds for every integer n≥1n \ge 1. This proof is non-circular: it uses only the definition of the Laplace transform and its existence theorem (assumed known, not derived from this result), plus ordinary calculus (integration by parts, limits). It does not invoke inverse transforms, convolution, or any result built on top of this theorem.

∎

The two special cases we need immediately follow by setting n=1n=1 and n=2n=2:

L{y′}(s)=sY(s)−y(0),L{y′′}(s)=s2Y(s)−sy(0)−y′(0),\mathcal{L}\{y'\}(s) = sY(s) - y(0), \qquad \mathcal{L}\{y''\}(s) = s^2Y(s) - sy(0) - y'(0),

where Y(s)=L{y(t)}(s)Y(s) = \mathcal{L}\{y(t)\}(s).

CorollaryAlgebraization of a linear constant-coefficient IVP

Let a,b,ca,b,c be real constants with a≠0a \ne 0, let g(t)g(t) be piecewise continuous on every finite interval and of exponential order, and consider the initial value problem

ay′′+by′+cy=g(t),y(0)=y0,y′(0)=y1.a y'' + b y' + c y = g(t), \qquad y(0) = y_0, \quad y'(0) = y_1.

Suppose the (unique, by the existence–uniqueness theorem for linear ODEs) solution y(t)y(t) and its derivatives y(t),y′(t)y(t), y'(t) are continuous and of exponential order, and y′′y'' is piecewise continuous — hypotheses automatically satisfied when gg is piecewise continuous and of exponential order, by standard theory of linear ODEs with constant coefficients. Let Y(s)=L{y(t)}(s)Y(s) = \mathcal{L}\{y(t)\}(s) and G(s)=L{g(t)}(s)G(s) = \mathcal{L}\{g(t)\}(s). Then for ss larger than the exponential order of yy,

a(s2Y(s)−sy0−y1)+b(sY(s)−y0)+cY(s)=G(s),a\big(s^2Y(s) - s y_0 - y_1\big) + b\big(sY(s) - y_0\big) + cY(s) = G(s),

equivalently

(as2+bs+c) Y(s)=G(s)+(as+b)y0+ay1.(as^2+bs+c)\,Y(s) = G(s) + (as+b)y_0 + a y_1.
Proof

Technique: direct substitution using the theorem, exploiting linearity of L\mathcal{L}. Apply the Laplace transform to both sides of ay′′+by′+cy=g(t)ay''+by'+cy=g(t). The transform is linear (a direct consequence of linearity of the integral, established in The Laplace Transform: Definition and Existence), so

a L{y′′}(s)+b L{y′}(s)+c L{y}(s)=L{g(t)}(s)=G(s).a\,\mathcal{L}\{y''\}(s) + b\,\mathcal{L}\{y'\}(s) + c\,\mathcal{L}\{y\}(s) = \mathcal{L}\{g(t)\}(s) = G(s).

By the theorem with n=2n=2 applied to yy (hypotheses satisfied as stated),

L{y′′}(s)=s2Y(s)−sy(0)−y′(0)=s2Y(s)−sy0−y1,\mathcal{L}\{y''\}(s) = s^2Y(s) - sy(0) - y'(0) = s^2Y(s) - sy_0 - y_1,

and by the theorem with n=1n=1,

L{y′}(s)=sY(s)−y(0)=sY(s)−y0.\mathcal{L}\{y'\}(s) = sY(s) - y(0) = sY(s) - y_0.

Substituting these into the transformed equation gives

a(s2Y(s)−sy0−y1)+b(sY(s)−y0)+cY(s)=G(s).a\big(s^2Y(s)-sy_0-y_1\big) + b\big(sY(s)-y_0\big) + cY(s) = G(s).

Collecting all terms containing Y(s)Y(s) on the left and all remaining terms (constants times known initial data, and G(s)G(s)) on the right:

(as2+bs+c)Y(s)−(as+b)y0−ay1=G(s)  ⟹  (as2+bs+c)Y(s)=G(s)+(as+b)y0+ay1.(as^2+bs+c)Y(s) - (as+b)y_0 - ay_1 = G(s) \;\Longrightarrow\; (as^2+bs+c)Y(s) = G(s) + (as+b)y_0 + ay_1.

Since a≠0a\ne 0, the polynomial as2+bs+cas^2+bs+c is not identically zero, so for all but finitely many values of ss (its roots) we may divide, giving

Y(s)=G(s)+(as+b)y0+ay1as2+bs+c.Y(s) = \frac{G(s) + (as+b)y_0 + ay_1}{as^2+bs+c}.

This is a purely algebraic equation for Y(s)Y(s): no derivatives, no integrals, no unknown functions besides Y(s)Y(s) itself. ■\blacksquare

∎

Worked examples

Example 1 (nonzero forcing, asymmetric initial data). Solve for Y(s)Y(s), given the IVP

y′′−3y′+2y=e4t,y(0)=2,y′(0)=−1.y'' - 3y' + 2y = e^{4t}, \qquad y(0) = 2, \quad y'(0) = -1.

Here a=1a=1, b=−3b=-3, c=2c=2, g(t)=e4tg(t)=e^{4t}. First, G(s)=L{e4t}(s)=1s−4G(s) = \mathcal{L}\{e^{4t}\}(s) = \dfrac{1}{s-4} for s>4s>4, a fact from the basic transform table established alongside the existence theorem. Transform both sides using linearity:

L{y′′}−3L{y′}+2L{y}=1s−4.\mathcal{L}\{y''\} - 3\mathcal{L}\{y'\} + 2\mathcal{L}\{y\} = \frac{1}{s-4}.

Substitute the derivative formulas with y(0)=2y(0)=2, y′(0)=−1y'(0)=-1:

L{y′′}=s2Y(s)−2s−(−1)=s2Y(s)−2s+1,\mathcal{L}\{y''\} = s^2Y(s) - 2s - (-1) = s^2Y(s) - 2s + 1,
L{y′}=sY(s)−2.\mathcal{L}\{y'\} = sY(s) - 2.

So the equation becomes

(s2Y(s)−2s+1)−3(sY(s)−2)+2Y(s)=1s−4.\big(s^2Y(s) - 2s + 1\big) - 3\big(sY(s)-2\big) + 2Y(s) = \frac{1}{s-4}.

Expand:

s2Y(s)−2s+1−3sY(s)+6+2Y(s)=1s−4.s^2Y(s) - 2s + 1 - 3sY(s) + 6 + 2Y(s) = \frac{1}{s-4}.

Group Y(s)Y(s) terms:

(s2−3s+2)Y(s)+(−2s+7)=1s−4.(s^2 - 3s + 2)Y(s) + (-2s + 7) = \frac{1}{s-4}.

Solve algebraically:

(s2−3s+2)Y(s)=1s−4+2s−7,(s^2-3s+2)Y(s) = \frac{1}{s-4} + 2s - 7,
Y(s)=1(s−4)(s2−3s+2)+2s−7s2−3s+2.Y(s) = \frac{1}{(s-4)(s^2-3s+2)} + \frac{2s-7}{s^2-3s+2}.

Factor the denominator s2−3s+2=(s−1)(s−2)s^2-3s+2=(s-1)(s-2), giving the fully algebraic answer

Y(s)=1(s−4)(s−1)(s−2)+2s−7(s−1)(s−2).\boxed{Y(s) = \frac{1}{(s-4)(s-1)(s-2)} + \frac{2s-7}{(s-1)(s-2)}}.

This rational function is exactly the input a partial-fraction decomposition (see Inverse Laplace Transform and Partial Fractions) would need in order to recover y(t)y(t); note we have not inverted anything here — the entire example is the algebraization step.

Example 2 (initial condition at a nonzero point via shifting is out of scope; instead: third-order flavor to test the general formula). Consider a third-order IVP to confirm the n=3n=3 case of the general theorem:

y′′′−2y′′+y′−2y=3t,y(0)=1, y′(0)=0, y′(0)(typo-check), y′′(0)=−2.y''' - 2y'' + y' - 2y = 3t, \qquad y(0)=1,\ y'(0)=0,\ y'(0)\text{(typo-check)}, \ y''(0)=-2.

To be precise, the data is y(0)=1y(0)=1, y′(0)=0y'(0)=0, y′′(0)=−2y''(0)=-2. By the theorem with n=3n=3:

L{y′′′}(s)=s3Y(s)−s2y(0)−s y′(0)−y′′(0)=s3Y(s)−s2−0⋅s−(−2)=s3Y(s)−s2+2.\mathcal{L}\{y'''\}(s) = s^3Y(s) - s^2 y(0) - s\,y'(0) - y''(0) = s^3Y(s) - s^2 - 0\cdot s - (-2) = s^3Y(s) - s^2 + 2.

Also

L{y′′}(s)=s2Y(s)−s⋅1−0=s2Y(s)−s,\mathcal{L}\{y''\}(s) = s^2Y(s) - s\cdot 1 - 0 = s^2Y(s) - s,
L{y′}(s)=sY(s)−1.\mathcal{L}\{y'\}(s) = sY(s) - 1.

The forcing term transforms as L{3t}(s)=3/s2\mathcal{L}\{3t\}(s) = 3/s^2 (s>0s>0), a standard entry from the basic table. Transform the full equation:

(s3Y−s2+2)−2(s2Y−s)+(sY−1)−2Y=3s2.\big(s^3Y - s^2+2\big) - 2\big(s^2Y - s\big) + \big(sY - 1\big) - 2Y = \frac{3}{s^2}.

Expand and collect Y(s)Y(s) terms:

s3Y−s2+2−2s2Y+2s+sY−1−2Y=3s2,s^3Y - s^2 + 2 - 2s^2Y + 2s + sY - 1 - 2Y = \frac{3}{s^2},
(s3−2s2+s−2)Y+(−s2+2s+1)=3s2.(s^3 - 2s^2 + s - 2)Y + (-s^2 + 2s + 1) = \frac{3}{s^2}.

Solve for Y(s)Y(s):

(s3−2s2+s−2) Y(s)=3s2+s2−2s−1,(s^3-2s^2+s-2)\,Y(s) = \frac{3}{s^2} + s^2 - 2s - 1,
Y(s)=1s3−2s2+s−2(3s2+s2−2s−1)=3+s4−2s3−s2s2(s3−2s2+s−2).Y(s) = \frac{1}{s^3-2s^2+s-2}\left(\frac{3}{s^2} + s^2-2s-1\right) = \frac{3 + s^4 - 2s^3 - s^2}{s^2(s^3-2s^2+s-2)}.

This is again a fully explicit algebraic (rational) expression for Y(s)Y(s), built purely from the coefficients 1,−2,1,−21,-2,1,-2, the forcing transform 3/s23/s^2, and the three initial values 1,0,−21,0,-2 — exactly as the general-order theorem predicts.

Counterexamplewhy continuity of ff (not just of f′f') is needed

Drop the hypothesis that ff itself is continuous on [0,∞)[0,\infty), keeping everything else (exponential order, piecewise continuity of the derivative). We show the naive formula L{f′}(s)=sF(s)−f(0)\mathcal{L}\{f'\}(s) = sF(s) - f(0) can fail — an extra jump term must be added. Let

f(t)={0,0≤t<11,t≥1.f(t) = \begin{cases} 0, & 0 \le t < 1 \\ 1, & t \ge 1. \end{cases}

This ff is bounded (hence of exponential order a=0a=0), and f′(t)=0f'(t) = 0 for all t≠1t \ne 1 (a piecewise continuous function, in fact identically zero except at the single point t=1t=1 where it is undefined — a measure-zero set that does not affect the integral). So L{f′}(s)=∫0∞e−st⋅0 dt=0\mathcal{L}\{f'\}(s) = \int_0^\infty e^{-st}\cdot 0 \, dt = 0. Now compute sF(s)−f(0)sF(s) - f(0). We have f(0)=0f(0) = 0, and

F(s)=∫0∞e−stf(t) dt=∫1∞e−st dt=e−ss(s>0).F(s) = \int_0^\infty e^{-st} f(t)\, dt = \int_1^\infty e^{-st}\, dt = \frac{e^{-s}}{s} \quad (s>0).

So

sF(s)−f(0)=s⋅e−ss−0=e−s.sF(s) - f(0) = s\cdot \frac{e^{-s}}{s} - 0 = e^{-s}.

Since e−s≠0e^{-s} \ne 0 for any real ss, we get

L{f′}(s)=0≠e−s=sF(s)−f(0)\mathcal{L}\{f'\}(s) = 0 \ne e^{-s} = sF(s) - f(0)

for every s>0s>0. The formula fails entirely, not just at isolated points. What went wrong: the proof's key step was the cancellation of interior boundary terms in integration by parts, and that cancellation relied on ff being continuous across the discontinuity of f′f' (at t=1t=1, where ff jumps from 00 to 11). Redoing the piecewise integration by parts explicitly, the correct identity picks up an extra term −e−s⋅1(f(1+)−f(1−))=−e−s(1−0)=−e−s-e^{-s\cdot 1}\big(f(1^+)-f(1^-)\big) = -e^{-s}(1-0) = -e^{-s}:

∫0∞e−stf′(t) dt=sF(s)−f(0)−e−s(1)=e−s−e−s=0,\int_0^\infty e^{-st}f'(t)\,dt = sF(s) - f(0) - e^{-s}(1) = e^{-s} - e^{-s} = 0,

which matches the direct computation L{f′}(s)=0\mathcal{L}\{f'\}(s)=0. This confirms that continuity of ff is not a technicality: without it, the clean formula sF(s)−f(0)sF(s)-f(0) must be corrected by a sum of jump terms, one for each discontinuity of ff, and the simple algebraization of an IVP breaks down unless the solution y(t)y(t) is genuinely continuous — which it always is for solutions of linear ODEs with continuous coefficients, but need not hold for an arbitrary function fed into the formula.

Scope, edge cases & common traps

SituationWhat happens
ff continuous, f′f' piecewise continuous, exponential orderClean formula L{f′}=sF(s)−f(0)\mathcal{L}\{f'\}=sF(s)-f(0) holds exactly as stated
ff itself discontinuous at t0t_0Extra term −e−st0[f(t0+)−f(t0−)]-e^{-st_0}[f(t_0^+)-f(t_0^-)] must be added (see counterexample)
Higher-order IVP, order nnNeed y,…,y(n−1)y,\dots,y^{(n-1)} continuous and of exponential order; y(n)y^{(n)} piecewise continuous
g(t)g(t) not of exponential order (e.g. et2e^{t^2})G(s)G(s) may fail to exist for any ss; the whole method breaks down before algebraization
Leading coefficient a=0a=0Corollary's algebraic step invalid; problem is not truly second order, formula for (as2+bs+c)(as^2+bs+c) degenerates
Initial conditions given at t=t0≠0t=t_0\ne 0Must shift variable to τ=t−t0\tau = t-t_0 first; the theorem as stated uses data at t=0t=0

The mathematician behind it

Origins · 1 minPierre-Simon Laplace and how this was discovered
PL
Pierre-Simon Laplace
1749–1827 · French mathematician, astronomer, and physicist working across probability, celestial mechanics, and mathematical physics in the era following Newton and Euler

Laplace's transform-like integral appeared first in his work on generating functions in probability theory and in his monumental Théorie analytique des probabilités and Mécanique céleste, where he needed systematic tools for solving linear difference and differential equations arising in orbital dynamics. The operational technique of converting differential equations into algebraic ones — using what we now call the Laplace transform — was fully developed only later, primarily by Oliver Heaviside in the 1890s for electrical circuit problems, and made rigorous still later by Bromwich, Doetsch, and others. Laplace's name attaches to the integral itself; the systematic use for solving IVPs, as presented here, is a later synthesis bearing his name in tribute to the foundational integral transform.

History
- 1782–1785: Laplace introduces integral transforms of the form ∫e−stf(t) dt\int e^{-st}f(t)\,dt in work on generating functions and probability.
- 1799–1825: Laplace applies transform-like methods extensively in Mécanique céleste to linear differential systems in celestial mechanics.
- 1892–1899: Oliver Heaviside develops an "operational calculus" for solving circuit ODEs, algebraically equivalent to today's Laplace-transform method, applying it to derivatives directly.
- 1916–1920: Thomas John I'Anson Bromwich supplies a rigorous complex-analytic foundation for Heaviside's operational methods, including the inversion integral.
- 1930s–1950s: Gustav Doetsch and others systematize the modern theory, formalizing exactly the hypotheses (piecewise continuity, exponential order) used in the theorem above.
Why does differentiation turn into multiplication by ss under the Laplace transform?

The mechanism is integration by parts applied to ∫0∞e−stf′(t) dt\int_0^\infty e^{-st}f'(t)\,dt. Since ddte−st=−se−st\frac{d}{dt}e^{-st} = -se^{-st}, integrating by parts moves the derivative off ff and onto the exponential kernel, producing a factor of ss times ∫0∞e−stf(t) dt=F(s)\int_0^\infty e^{-st}f(t)\,dt = F(s), plus a boundary term evaluated at t=0t=0 (the term at t=∞t=\infty vanishes because ff has exponential order less than ss). The boundary term at t=0t=0 is exactly −f(0)-f(0). So the "ss" arises purely from differentiating the kernel e−ste^{-st}, not from any property of ff itself — this is why the rule works for essentially any function of exponential order with a piecewise continuous derivative.

Do I need to check exponential order every time I use this formula in practice?

In a first course, no — most functions in textbook IVPs (polynomials, exponentials, sines, cosines, and their products) are automatically of exponential order, and this is usually stated once and then assumed. Rigorously, though, the hypothesis is essential: it guarantees both that F(s)F(s) exists for large ss and that the boundary term e−sAf(A)→0e^{-sA}f(A)\to 0 in the proof. Functions like et2e^{t^2} grow too fast and have no Laplace transform for any ss, so the whole method fails before you even get to differentiate. In practice, check exponential order once for the forcing function g(t)g(t); polynomial, exponential, and bounded-oscillatory solutions of linear ODEs are always of exponential order.

What if the initial conditions are given at t=t0≠0t=t_0\ne 0 instead of t=0t=0?

The theorem as stated requires initial data at t=0t=0, since the Laplace transform integral itself starts at t=0t=0. If conditions are given at t=t0t=t_0, substitute τ=t−t0\tau = t - t_0 to shift the problem so that the new independent variable starts at 00, then apply the theorem to the shifted function y~(τ)=y(τ+t0)\tilde y(\tau) = y(\tau+t_0). The transform machinery does not directly handle nonzero initial points without this substitution; attempting to plug y(t0)y(t_0) directly into the formula in place of y(0)y(0) is a common error that produces an incorrect algebraic equation.

Does this method work for variable-coefficient linear ODEs, like ty′′+y′+ty=0ty''+y'+ty=0 (Bessel's equation)?

Sometimes, but the algebraization is more subtle. Multiplying by tt corresponds to differentiating the transform with respect to ss (a companion rule, L{tf(t)}=−F′(s)\mathcal{L}\{tf(t)\}=-F'(s)), so an equation with a tt-coefficient becomes a differential equation in ss rather than a purely algebraic one. The clean algebraic equation in this article is special to constant coefficients; variable-coefficient equations trade one differential equation for another, sometimes simpler, differential equation in the transform variable.

Why do we subtract f(0)f(0) and not, say, f(∞)f(\infty) or some other value?

The subtracted term comes directly from evaluating the antiderivative boundary [e−stf(t)]0∞\big[e^{-st}f(t)\big]_0^\infty in the integration-by-parts computation. The upper limit contributes lim⁡A→∞e−sAf(A)\lim_{A\to\infty}e^{-sA}f(A), which is forced to be 00 precisely because ss exceeds the exponential order of ff — this is why the theorem requires s>as>a. Only the lower limit, at t=0t=0, survives, contributing −e0f(0)=−f(0)-e^{0}f(0)=-f(0). If the transform were instead defined as a two-sided (bilateral) transform over all of R\mathbb{R}, there would be no boundary term at all — this is a special feature of the one-sided transform matched to initial value problems.

Can this method handle systems of ODEs, not just a single second-order equation?

Yes. Apply the same derivative rule to each equation in the system separately, transforming each unknown function y1(t),y2(t),…y_1(t), y_2(t),\dots into Y1(s),Y2(s),…Y_1(s), Y_2(s),\dots. The result is a linear system of algebraic equations in the transformed variables, which can be solved by ordinary linear algebra (elimination, Cramer's rule, or matrix inversion) exactly as a single equation is solved above. This is one of the main reasons the method scales well to coupled spring–mass systems and multi-loop circuits.

What comes after finding Y(s)Y(s) — how do I actually get back y(t)y(t)?

Finding Y(s)Y(s), as done throughout this article, is only the algebra step. To recover y(t)y(t) you invert the transform, typically by recognizing Y(s)Y(s) as a sum of standard transform-table entries after a partial fraction decomposition of the rational function in ss. That inversion process, including how to handle repeated roots, complex conjugate roots, and forcing terms combined via convolution, is developed in Inverse Laplace Transform and Partial Fractions and Step Functions and Convolution.

Before this

Where this leads