Laplace Transform in Time¶

Everything in this subsection follows from one sentence:

The Laplace transform turns the time derivative \(\partial_\tau\) into multiplication by \(p\).

Where Fourier and Mellin diagonalize the spatial operator in the Black–Scholes PDE and leave an ODE in time, the Laplace transform does the opposite — it freezes time and leaves an ODE in \(S\). For the constant-coefficient Black–Scholes operator that ODE is an Euler–Cauchy equation with power-law solutions \(S^\lambda\). Inversion (the Bromwich integral) reassembles \(V(S, \tau)\) by integrating contributions from every \(p\).

We build the picture in the same order a reader naturally builds intuition: a toy first-order ODE in \(\tau\), then the BS PDE, and only at the end the operator-theoretic interpretation that names what we have done a resolvent calculation.

A Toy Time-Evolution Problem¶

Before any finance, work the simplest possible time-evolution problem:

— a first-order ODE with constant coefficient \(a\) and known initial datum. Its solution is of course \(u(\tau) = u_0\, e^{-a\tau}\), but deriving it through the Laplace transform demonstrates the mechanism that will lift to the BS PDE essentially unchanged.

The Mechanism¶

Apply the Laplace transform \(\tilde u(p) = \int_0^\infty e^{-p\tau} u(\tau)\, d\tau\) to both sides. The defining identity of the Laplace transform — the only one we will need — is

Time differentiation becomes multiplication by \(p\), minus the initial value. Substituting:

The differential equation has become a single algebraic equation in \(p\). To recover \(u(\tau)\), invert via the Bromwich integral:

closing the contour around the simple pole at \(p = -a\) and reading off the residue. The decay rate of the original time-evolution problem is the location of the pole of its Laplace transform.

What Carries Forward¶

Two observations lift directly to the BS PDE without modification:

• Pole structure encodes time decay. A simple pole at \(p = -a\) corresponds to an exponential \(e^{-a\tau}\). The inverse Laplace transform — sum of residues, or a Bromwich contour — is the spectral decomposition of the time-evolution operator. When we lift to a PDE, the algebraic poles of the toy problem become branch points and resolvent singularities of the spatial operator.
• The coefficient \(a\) becomes a spatial operator. In the toy ODE, the right-hand side is \(-a\, u\) with \(a\) a number. In the BS PDE, the right-hand side is \(\mathcal{L} V\) with \(\mathcal{L}\) a differential operator in \(S\). Laplace transforming still gives an "algebraic" equation, but it is now an ODE in \(S\) — one ODE for every value of \(p\).

Why Laplace and Not Fourier¶

Two reasons the time-axis problem \(u' = -a u\) wants Laplace, not Fourier:

• Causality. The problem is set on \([0, \infty)\) with initial data at \(\tau = 0\), not on the full real line. Laplace integrates over \([0, \infty)\) and treats the initial condition explicitly via the \(-u(0)\) term in the differentiation identity. Fourier on the half-line would require an artificial extension and lose the initial condition as a clean object.
• Exponential growth is allowed. \(u(\tau) = u_0 e^{a\tau}\) (\(a > 0\)) is fine for Laplace as long as \(\operatorname{Re}(p) > a\); Fourier would require boundedness, which \(u\) lacks.

Whenever the time axis is one-sided and the initial condition is the natural data, Laplace is the right transform. Black–Scholes time-to-maturity \(\tau \in [0, T]\) is exactly that setup.

Spatial Modes and Temporal Decay¶

Fourier and Laplace are not competitors but complementary halves of one picture. The cleanest illustration is the heat equation \(u_t = \kappa\, u_{xx}\). Decompose the spatial profile \(u(\cdot, t)\) into Fourier modes:

Substituting one mode into the PDE gives \(\hat u_t(\omega, t) = -\kappa\omega^2\, \hat u(\omega, t)\) — a first-order ODE in \(\tau\) of exactly the form \(u' = -a u\) from the toy above, with decay rate \(a = \kappa\omega^2\). Solving:

This is the spectral-semigroup picture: oscillation in space, exponential decay in time. Fourier decomposes space into eigenmodes \(e^{i\omega x}\); each mode evolves by multiplication by an exponential whose rate depends on the wavenumber; Laplace is the natural transform of those exponentials. High frequencies decay faster (\(\omega^2\) scaling) — which is exactly why heat smooths sharp features.

For Black–Scholes the same picture holds verbatim, with log-price coordinates supplying the spatial modes:

• Fourier/Mellin \(\to\) spatial modes (plane waves \(e^{i\omega x}\) or power laws \(S^s\));
• Time evolution \(\to\) each mode multiplied by an exponential \(e^{\psi(\omega)\tau}\) (characteristic exponent of the diffusion);
• Laplace \(\to\) treats those exponentials algebraically via poles and resolvents — the eigenvalues of the time-translation semigroup.

The kernel duality is exact: writing \(p = i\omega\) in the Bromwich integral recovers the Fourier transform, so

Laplace handles what Fourier cannot — exponentially decaying or growing initial data, one-sided time axes, transient pole structure — while sharing the same spectral decomposition once \(\operatorname{Re}(p) > 0\). The full Combined Fourier–Laplace section below makes the two-variable factorisation \(\tilde{\hat V}(\omega, p) = \hat\Phi(\omega) / (p - \psi(\omega))\) explicit.

This is the mechanism. Black–Scholes is the application.

Laplace Transform Definition¶

The Laplace transform of \(V(S,\tau)\) in time-to-maturity \(\tau = T - t\) is

with inverse given by the Bromwich integral

where \(c\) lies to the right of all singularities of \(\tilde{V}\). The key operational property is

with \(V(S,0) = \Phi(S)\) the payoff — this is what converts time differentiation into multiplication by \(p\).

Transforming the Black-Scholes PDE¶

Because \(S\) is not the transform variable, \(S\)-derivatives pass through the integral. Applying the Laplace transform in \(\tau\) and using \(\mathcal{L}[\partial_\tau V] = p\tilde{V} - \Phi(S)\) turns the PDE into the resolvent equation

a second-order ODE in \(S\) — the time variable has been completely eliminated. This is the mirror of Fourier/Mellin, which eliminate the spatial variable and leave an ODE in time.

The Euler-Cauchy Structure¶

The homogeneous part of the transformed equation,

is an Euler-Cauchy (equidimensional) equation. Its solutions are power laws \(\tilde{V} = S^{\lambda}\). Substituting this ansatz:

Expanding:

This is a quadratic in \(\lambda\) with roots

For \(\operatorname{Re}(p)\) sufficiently large, the discriminant is positive and the two roots are real with \(\lambda_+ > 0\) and \(\lambda_- < 0\).

The general homogeneous solution is therefore

where \(A(p)\) and \(B(p)\) are determined by boundary conditions and the payoff.

Solving for the European Call¶

For \(\Phi(S) = (S - K)^+\), write \(\tilde{V}(S,p) = A(p) S^{\lambda_+} + B(p) S^{\lambda_-} + \tilde{V}_p(S,p)\), with \(\tilde{V}_p\) a particular solution built from the Euler–Cauchy Green's function. Boundary conditions \(\tilde{V} \to 0\) as \(S \to 0\) and \(\tilde{V} \sim S/p\) as \(S \to \infty\) fix the constants. Bromwich inversion then recovers the Black–Scholes formula; the \((d_1, d_2)\) structure is the canonical § Heat Equation derivation and is not reproved here.

Log-Price Formulation¶

In log-price coordinates \(x = \ln S\) — the same change of variables that converts Black–Scholes to the § Heat Equation — the resolvent ODE becomes

a constant-coefficient ODE (the Euler–Cauchy \(S\)-dependence is absorbed). The characteristic quadratic is unchanged and the homogeneous solutions are \(e^{\lambda_\pm x}\).

The Green's function on \(x \in (-\infty, \infty)\) is

and the particular solution is \(\tilde{V}(x,p) = \int_{-\infty}^{\infty} \tilde{G}(x,y;p) \, \Phi(e^y) \, dy\).

Laplace Inversion and Numerical Methods¶

The option price is recovered from \(\tilde{V}(S,p)\) via the Bromwich integral

In practice, numerical inversion is the working tool. The standard methods are the Gaver–Stehfest algorithm (real evaluations only, limited precision), Talbot's method (deformed contour, rapid convergence), and the Abate–Whitt Euler method (trapezoidal rule with Euler summation). For Black–Scholes with constant parameters, Gaver–Stehfest with \(N = 12\) typically gives 5–6 digits.

Combined Fourier–Laplace Transform¶

Applying Fourier in \(x\) and Laplace in \(\tau\) together gives a complete spectral decomposition: Fourier diagonalises the spatial generator \(\mathcal{L}\) in \(\omega\)-space (each plane wave \(e^{i\omega x}\) is an eigenfunction with eigenvalue \(\psi(\omega)\) — see § Fourier Transform), while Laplace diagonalises the time-translation semigroup in \(p\)-space.

The double transform reduces the PDE to the algebraic equation \((p - \psi(\omega))\tilde{\hat{V}}(\omega,p) = \hat{\Phi}(\omega)\) with characteristic exponent \(\psi(\omega) = -\tfrac{\sigma^2 \omega^2}{2} + i\omega(r - \tfrac{\sigma^2}{2}) - r\), so

Inverting first in \(p\) (simple pole at \(p = \psi(\omega)\)) yields \(\hat{V}(\omega,\tau) = \hat{\Phi}(\omega)e^{\psi(\omega)\tau}\), which is the § Fourier Transform solution. The two inversions commute, and the resolvent factorisation makes the consistency manifest. The subsequent inversion in \(\omega\) that recovers the heat-kernel representation is the canonical § Heat Equation calculation; we do not repeat it here.

Geometrically: \(\omega\) parametrises spatial modes, \(p\) parametrises temporal modes, and the pole locus \(p = \psi(\omega)\) is the dispersion relation of the pricing operator. This is the same structure as in the Schrödinger / heat-equation literature, transplanted to finance.

The Semigroup-Resolvent Identity¶

The Laplace transform and the pricing semigroup are linked by a single identity that unifies semigroup theory, Green's functions, and Laplace inversion:

Read both sides as operators on payoffs. The right-hand side is the Laplace transform of the pricing semigroup \(\mathcal{P}_\tau = e^{\tau\mathcal{L}}\) (recall § Introduction § A Unifying View); the left-hand side is the resolvent \(R_p := (pI - \mathcal{L})^{-1}\) of the generator \(\mathcal{L}\). The identity holds whenever \(\operatorname{Re}(p) > \omega(\mathcal{L})\) (the growth bound of the semigroup), and the integral converges in the operator norm. It is the operator-valued version of the elementary scalar identity \(1/(p - a) = \int_0^\infty e^{-p\tau}\, e^{a\tau}\, d\tau\) that drove the toy ODE above.

Three of the chapter's representations collapse onto this one identity:

• Semigroup view. \(e^{\tau\mathcal{L}}\) is the pricing operator; its Laplace transform is the resolvent.
• Green's-function view. Applied to a payoff \(\Phi\), the right-hand side is \(\int_0^\infty e^{-p\tau}\, V(\cdot, \tau)\, d\tau = \tilde V(\cdot, p)\) — the Laplace transform of the price — and the left-hand side is the Green's function of the modified operator \((pI - \mathcal{L})\), namely the time-independent BS PDE with shift \(-p\). The Euler–Cauchy structure of §"The Euler-Cauchy Structure" is exactly the resolvent of \(\mathcal{L}\) acting on power-law payoffs.
• Spectral view. The poles of \(R_p\) on the real axis are the (generalised) eigenvalues of \(\mathcal{L}\); the Bromwich integral that inverts \(\tilde V(\cdot, p)\) is the spectral decomposition of the pricing semigroup. For pure BS, the spectrum is continuous (along the parabola \(p = \psi(\omega)\)); for bounded-domain problems (barrier options, § Separation of Variables) it becomes discrete.

The resolvent identity is therefore the bridge between "solve the time-independent ODE in \(S\) at each \(p\)" (the calculation we have been doing) and the abstract operator picture in which Black–Scholes pricing is Laplace inversion of the resolvent of a parabolic generator. The perpetual-options limit below is the \(p = 0\) case: the resolvent at zero, \(\mathcal{L}^{-1}\), is exactly the Green's function of the time-independent BS PDE.

Perpetual Options as the Zero-Frequency Limit¶

The Laplace transform connects naturally to perpetual (infinite-maturity) options. For a perpetual option, \(\partial_\tau V = 0\) and the PDE reduces to

which is exactly the homogeneous Euler–Cauchy equation at \(p = 0\). The Laplace parameter \(p\) thus acts as a frequency conjugate to time: \(p = 0\) is the stationary (perpetual) case; \(p \neq 0\) captures transient decay.

For a perpetual American put with strike \(K\), smooth pasting at \(S^*\) (\(V(S^*) = K - S^*\), \(V'(S^*) = -1\)) and \(V \to 0\) as \(S \to \infty\) select the \(\lambda_-(0)\) branch, giving \(V(S) = (K - S^*)(S/S^*)^{\lambda_-}\) for \(S > S^*\) (see Exercise 5).

Where Laplace Genuinely Helps¶

The two settings where the Laplace approach dominates Fourier are worth naming explicitly.

• Time-dependent coefficients. When \(\sigma = \sigma(t)\) or \(r = r(t)\), the Fourier-only approach fails because the characteristic exponent itself becomes time-dependent and the ODE in \(\tau\) no longer has exponential solutions. But the Laplace transform in \(\tau\) still removes the time derivative, leaving an ODE in \(S\) whose coefficients depend on \(p\) in a controlled way. A concrete case: deterministic volatility \(\sigma(t)\) entering through an integrated variance \(\Sigma(\tau) = \int_0^\tau \sigma^2(s)\,ds\). A change of time variable \(\tau \mapsto \Sigma\) followed by Laplace transform reduces the problem to the constant-coefficient case studied above, with \(p\) as the conjugate variable.

• First-passage and barrier problems. For barrier and stopping-time problems the Laplace transform of the hitting-time distribution is naturally a power \(S^{\lambda_\pm(p)}\) — the same Euler–Cauchy roots reappear. The perpetual put above is in fact the simplest such example (with \(p = 0\)). For Parisian and occupation-time options, the Laplace framework is essentially the only tractable one.

In short: whenever the time structure resists separation of variables, Laplace turns the obstruction into an algebraic dependence on \(p\).

Advantages and Limitations¶

For constant-parameter European options the Fourier route with FFT is typically faster; numerical Laplace inversion is inherently ill-conditioned. Laplace earns its place on problems where the time structure resists separation: time-dependent parameters, American/free-boundary problems, and perpetual structures.

The unifying picture is the resolvent identity \(\tilde{V}(S,p) = (p - \mathcal{L})^{-1}\Phi\). Inversion via the Bromwich integral is the operator-theoretic statement \(\mathcal{P}_\tau = \frac{1}{2\pi i} \oint e^{p\tau}(p - \mathcal{L})^{-1} dp\), expressing the semigroup as a contour integral of its resolvent — exactly the Hille–Yosida / Dunford functional calculus, specialised to the Black–Scholes generator.

Exercises¶

Exercise 1. Starting from the log-price Black-Scholes PDE, apply the Laplace transform in the time variable \(\tau\) and derive the resulting constant-coefficient ODE in \(x = \ln S\). Find the characteristic equation and its roots \(\lambda_{\pm}(p)\). Verify that \(\lambda_+(p) > 0 > \lambda_-(p)\) when \(p > 0\).

Exercise 2. The Euler-Cauchy equation \(\frac{\sigma^2}{2}S^2 \tilde{V}'' + rS\tilde{V}' - (r+p)\tilde{V} = 0\) has solutions \(S^{\lambda_{\pm}}\). Verify this by direct substitution. Then show that the product \(\lambda_+ \cdot \lambda_-\) equals \(-2(r+p)/\sigma^2\) and explain why this implies the two solutions have different growth behavior as \(S \to \infty\).

Exercise 3. For the combined Fourier-Laplace transform, show that the double transform \(\tilde{\hat{V}}(\omega,p) = \hat{\Phi}(\omega)/(p - \psi(\omega))\) has a simple pole in the \(p\)-variable at \(p = \psi(\omega)\). Evaluate the Bromwich integral in \(p\) using the residue theorem and verify that one recovers the standard Fourier-only solution \(\hat{V}(\omega,\tau) = \hat{\Phi}(\omega)e^{\psi(\omega)\tau}\).

Exercise 4. The Gaver-Stehfest method approximates the inverse Laplace transform using only real-valued evaluations: \(V(S,\tau) \approx \frac{\ln 2}{\tau}\sum_{k=1}^{N} c_k \, \tilde{V}(S, k\ln 2/\tau)\). Explain why this method requires high-precision arithmetic when \(N\) is large. For the Black-Scholes call in Laplace space, describe how you would compute \(\tilde{V}(S,p)\) for a given real \(p > 0\).

Exercise 5. Show that the Laplace transform of the Black-Scholes PDE with \(p = 0\) recovers the ODE for perpetual options. For a perpetual American put with strike \(K\), use the boundary conditions \(V(S^*) = K - S^*\) and \(V'(S^*) = -1\) (smooth pasting) together with \(V(S) \to 0\) as \(S \to \infty\) to find \(S^*\) and \(V(S)\) in closed form.

Exercise 6. The singularity structure of \(\tilde{V}(S,p)\) in the complex \(p\)-plane determines the large-\(\tau\) behavior of \(V(S,\tau)\). The discriminant of the characteristic equation vanishes at \(p^* = -r - \frac{(r - \sigma^2/2)^2}{2\sigma^2}\). Explain why \(p^*\) is a branch point of \(\tilde{V}(S,p)\), and show that for \(\tau \to \infty\) the option price decays as \(V(S,\tau) \sim e^{p^* \tau} \tau^{-1/2}\) (up to algebraic prefactors).