Uniqueness via Energy Methods¶
The maximum principle provides one route to uniqueness for the heat equation. Energy methods offer a complementary approach that is often more flexible: define an integral quantity (the "energy") that measures how far a solution is from zero, and show this energy can only decrease over time. If the energy starts at zero, it must remain zero -- proving the solution is identically zero and hence unique.
Related Content
- Maximum Principle -- the other main uniqueness tool
- Fundamental Solution -- the heat kernel
- Heat Equation and Brownian Motion -- probabilistic interpretation
The Energy Functional¶
Definition¶
For a function \(u(t, x)\) on the domain \(Q_T = (0, T] \times (a, b)\), define the energy:
This is a non-negative quantity that measures the "total magnitude" of \(u\) at time \(t\). It equals zero if and only if \(u(t, \cdot) = 0\) almost everywhere.
Physical Intuition¶
The energy \(E(t)\) has a natural interpretation:
- In heat conduction: \(E(t)\) is proportional to the total thermal energy stored in the rod
- In probability: If \(u\) is a density perturbation, \(E(t)\) measures the \(L^2\) distance from zero
- In finance: For the difference of two portfolio values, \(E(t)\) measures the total squared discrepancy
The heat equation dissipates energy: diffusion spreads out concentrations, reducing peaks and filling valleys. This smoothing drives \(E(t)\) down.
Energy Dissipation for the Heat Equation¶
Theorem (Energy Decay)¶
Let \(u\) solve the heat equation on \((0, T] \times (a, b)\):
with homogeneous Dirichlet boundary conditions \(u(t, a) = u(t, b) = 0\). Then:
The energy is non-increasing. It is strictly decreasing unless \(u_x \equiv 0\), which (combined with the boundary conditions) forces \(u \equiv 0\).
Proof¶
Differentiate under the integral sign (justified by the smoothness of solutions to the heat equation):
Substitute the heat equation \(u_t = \frac{1}{2}u_{xx}\):
Integrate by parts: Using \(\int u\,u_{xx}\,dx = [u\,u_x]_a^b - \int u_x^2\,dx\):
The boundary term vanishes because \(u(t, a) = u(t, b) = 0\):
\(\square\)
The Key Step: Integration by Parts
The integration by parts transfers one derivative from \(u_{xx}\) to \(u\), producing the squared gradient \(u_x^2\). This is the essential trick in energy methods: convert a product involving second derivatives into a non-negative squared quantity.
Uniqueness Proof¶
Theorem (Uniqueness via Energy)¶
The initial-boundary value problem:
has at most one solution in \(C^{2,1}(\overline{Q}_T)\).
Proof¶
Suppose \(u_1\) and \(u_2\) are two solutions with the same data. Let \(w = u_1 - u_2\). Then \(w\) satisfies:
Define the energy of the difference:
Step 1: At \(t = 0\), since \(w(0, x) = 0\):
Step 2: By the energy decay result:
Step 3: Since \(E(t) \geq 0\) always, and \(E(0) = 0\), and \(E\) is non-increasing:
Step 4: \(E(t) = 0\) implies \(\int_a^b w^2\,dx = 0\), hence \(w(t, x) = 0\) for all \(x \in [a, b]\).
Therefore \(u_1 = u_2\). \(\square\)
Sharper Energy Estimates¶
Poincare Inequality¶
On the interval \((a, b)\) with homogeneous Dirichlet conditions, the Poincare inequality states:
Combining with the energy dissipation:
This is a Gronwall-type inequality, yielding exponential decay:
Interpretation: The energy decays exponentially with rate \(\pi^2 / (b-a)^2\). Shorter intervals (smaller \(b - a\)) lead to faster decay -- the diffusion reaches the boundaries sooner.
Connection to Spectral Theory¶
The decay rate \(\pi^2 / (b-a)^2\) is the first eigenvalue of \(-\frac{1}{2}\partial_{xx}\) on \((a, b)\) with Dirichlet conditions. The slowest-decaying mode is the fundamental eigenfunction \(\sin\!\left(\frac{\pi(x-a)}{b-a}\right)\), which decays at exactly this rate. All higher modes decay faster.
Extension to Variable Coefficients¶
General Parabolic Operator¶
Consider the variable-coefficient equation:
where \(\sigma^2(x) \geq \sigma_{\min}^2 > 0\) (uniform ellipticity).
The energy dissipation becomes:
Uniqueness follows by the same argument.
With Lower-Order Terms¶
For the equation \(u_t = \frac{1}{2}u_{xx} + \mu(x)u_x - r(x)u\) with \(r \geq 0\), define the weighted energy:
Choosing \(\lambda\) large enough to absorb the drift term, one obtains \(\frac{d\tilde{E}}{dt} \leq 0\), yielding uniqueness.
Financial Application
For the Black-Scholes PDE with drift \(rS\partial_S\) and discounting \(-rV\), the energy method applies after a change of variables \(x = \log S\), which transforms the equation into one with bounded coefficients on \(\mathbb{R}\).
Neumann Boundary Conditions¶
For Neumann conditions \(u_x(t, a) = u_x(t, b) = 0\), the boundary term in the integration by parts is:
The energy dissipation \(\frac{dE}{dt} = -\frac{1}{2}\int u_x^2\,dx \leq 0\) still holds. However, the Poincare inequality must be modified: for Neumann conditions, the constant function is in the kernel, so uniqueness holds only up to a constant.
Energy Methods vs Maximum Principle¶
| Aspect | Energy Methods | Maximum Principle |
|---|---|---|
| What it bounds | \(L^2\) norm (\(\int u^2\,dx\)) | \(L^\infty\) norm ($\max |
| Proof technique | Integration by parts | Calculus of maxima |
| Extends to | Higher dimensions, systems | Scalar equations primarily |
| Quantitative | Gives decay rate | Gives pointwise bounds |
| Regularity needed | Weaker (\(H^1\)) | Stronger (\(C^2\)) |
| Variable coefficients | Natural extension | Needs uniform ellipticity |
Complementary Tools
Neither method strictly dominates the other. The maximum principle gives sharper pointwise bounds but is harder to extend to systems. Energy methods are more flexible and extend naturally to weak (Sobolev) solutions, making them the primary tool in modern PDE theory.
The Probabilistic Connection¶
Energy dissipation has a probabilistic interpretation. If \(u(t, x)\) represents the density of a perturbation to Brownian motion, then:
decreases because Brownian motion mixes: it spreads out any initial concentration. The rate of mixing is governed by the spectral gap (the first eigenvalue), which determines how quickly the process "forgets" its initial condition.
In the language of probability: the transition semigroup is contractive in \(L^2\).
Summary¶
| Result | Statement |
|---|---|
| Energy decay | \(E(t) \leq E(0)\) for all \(t > 0\) |
| Uniqueness | \(E(0) = 0 \implies E(t) = 0\) for all \(t\) |
| Exponential decay | \(E(t) \leq E(0)\,e^{-\pi^2 t/(b-a)^2}\) with Poincare |
| Variable coefficients | Holds with \(\sigma_{\min}^2 > 0\) |
Energy methods prove uniqueness by showing that diffusion dissipates the total squared magnitude of any solution. The energy can only go down, so a solution starting at zero must remain at zero.
See Also¶
- Maximum Principle -- complementary uniqueness tool
- Fundamental Solution -- explicit solution showing decay
- Spectral Decomposition -- eigenfunction expansion and decay rates
- Scaling and Invariance -- self-similar structure of the heat equation
Exercises¶
Exercise 1. Define the energy functional \(E(t) = \frac{1}{2}\int_a^b u(x, t)^2\,dx\) for a solution \(u\) of the heat equation with homogeneous Dirichlet conditions on \([a, b]\). Compute \(E'(t)\) and show that \(E'(t) \leq 0\), proving that energy is non-increasing.
Solution to Exercise 1
Define \(E(t) = \frac{1}{2}\int_a^b u(x,t)^2\,dx\) where \(u\) solves \(\partial_t u = \frac{1}{2}\partial_{xx}u\) with \(u(t,a) = u(t,b) = 0\).
Differentiating under the integral sign:
Integrate by parts: \(\int_a^b u\,u_{xx}\,dx = [u\,u_x]_a^b - \int_a^b u_x^2\,dx\).
The boundary term vanishes because \(u(t,a) = u(t,b) = 0\):
Therefore:
since \(u_x^2 \geq 0\). The energy is non-increasing. It is strictly decreasing unless \(u_x \equiv 0\) on \([a,b]\), which combined with \(u(a) = u(b) = 0\) forces \(u \equiv 0\).
Exercise 2. Using the energy method, prove uniqueness: if \(u\) and \(v\) both solve the heat equation with the same initial and boundary data, define \(w = u - v\) and show \(E_w(t) = 0\) for all \(t \geq 0\), hence \(w = 0\).
Solution to Exercise 2
Let \(u\) and \(v\) both solve the heat equation with the same initial data \(u(0,x) = v(0,x) = f(x)\) and boundary data \(u(t,a) = v(t,a)\), \(u(t,b) = v(t,b)\).
Define \(w = u - v\). By linearity, \(w\) solves:
Define the energy of the difference:
Step 1: \(E_w(0) = \frac{1}{2}\int_a^b 0^2\,dx = 0\).
Step 2: By the energy decay result (Exercise 1), \(E_w'(t) \leq 0\) for all \(t > 0\).
Step 3: Since \(E_w(t) \geq 0\) (it is an integral of a squared function) and \(E_w(0) = 0\) and \(E_w\) is non-increasing, we must have \(E_w(t) = 0\) for all \(t \geq 0\).
Step 4: \(E_w(t) = 0\) means \(\int_a^b w^2\,dx = 0\). Since \(w^2 \geq 0\) and continuous, this implies \(w(x,t) = 0\) for all \(x \in [a,b]\).
Therefore \(u = v\) everywhere on \([a,b] \times [0,T]\).
Exercise 3. The Poincare inequality states \(\int_a^b u^2\,dx \leq C\int_a^b (u')^2\,dx\) for functions vanishing at the endpoints, with \(C = (b-a)^2/\pi^2\). Use this together with the energy decay \(E'(t) = -\int (u')^2\,dx\) to show \(E'(t) \leq -\frac{\pi^2}{(b-a)^2}E(t)\), giving exponential decay \(E(t) \leq E(0)e^{-\pi^2 t/(b-a)^2}\).
Solution to Exercise 3
From Exercise 1, \(E'(t) = -\frac{1}{2}\int_a^b u_x^2\,dx\). The Poincare inequality for functions vanishing at the endpoints states:
Rearranging: \(\int_a^b u_x^2\,dx \geq \frac{\pi^2}{(b-a)^2}\int_a^b u^2\,dx = \frac{2\pi^2}{(b-a)^2}E(t)\).
Substituting into the energy decay formula:
This is the differential inequality \(E'(t) \leq -\mu E(t)\) with \(\mu = \pi^2/(b-a)^2\). By Gronwall's inequality (or direct comparison with \(E(0)e^{-\mu t}\)):
The energy decays exponentially with rate \(\pi^2/(b-a)^2\). This rate is the first eigenvalue of \(-\frac{d^2}{dx^2}\) on \((a,b)\) with Dirichlet conditions, corresponding to the slowest-decaying Fourier mode \(\sin(\pi(x-a)/(b-a))\).
Exercise 4. For Neumann conditions \(u_x(a, t) = u_x(b, t) = 0\), the energy \(E(t) = \frac{1}{2}\int u^2\,dx\) still decreases, but does it decay to zero? Explain why the Poincare inequality must be modified (applied to \(u - \bar{u}\) where \(\bar{u}\) is the spatial average) and what the equilibrium solution is.
Solution to Exercise 4
For Neumann conditions \(u_x(a,t) = u_x(b,t) = 0\), the energy dissipation still holds:
since the boundary term \(u(b)u_x(b) - u(a)u_x(a) = 0\) by the Neumann conditions.
However, \(E(t)\) does not decay to zero in general. The constant function \(u(x,t) = c\) satisfies the heat equation, \(u_x = 0\) (so the Neumann conditions hold), and has energy \(E = c^2(b-a)/2 \neq 0\). This is the equilibrium solution.
The Poincare inequality must be modified: for Neumann conditions, one uses the Poincare-Wirtinger inequality:
where \(\bar{u} = \frac{1}{b-a}\int_a^b u\,dx\) is the spatial average. Note that \(\bar{u}\) is conserved in time (by integrating the heat equation over \([a,b]\) and using Neumann conditions). The deviation \(u - \bar{u}\) decays exponentially to zero, so:
The equilibrium is the spatial average of the initial data, representing uniform temperature.
Exercise 5. Explain the financial interpretation of energy decay for barrier option pricing on a bounded domain \([B_l, B_u]\). As \(T - t \to \infty\), the option price decays exponentially. Relate the decay rate to the smallest eigenvalue \(\lambda_1 = \pi^2 / (2(B_u - B_l)^2)\).
Solution to Exercise 5
In barrier option pricing on a bounded domain \([B_l, B_u]\) (lower and upper barriers), the option price \(V\) satisfies the heat equation (after appropriate transformations) with Dirichlet conditions \(V = 0\) at both barriers (the option is knocked out).
By the energy method with the Poincare inequality, the \(L^2\) energy of \(V\) decays as:
where \(\lambda_1 = \frac{\pi^2}{2(B_u - B_l)^2}\) is the first eigenvalue of the operator \(-\frac{1}{2}\partial_{xx}\) on the interval with Dirichlet conditions.
Financial interpretation: As \(T - t \to \infty\) (very long time to expiry), the probability that Brownian motion has not hit either barrier becomes exponentially small, dominated by the slowest-decaying eigenmode. The option price decays exponentially with rate \(\lambda_1\) because the particle (stock price) is almost certain to be knocked out.
The decay rate is faster for narrower barrier intervals (smaller \(B_u - B_l\)): the closer the barriers, the sooner the stock price hits one of them, and the faster the option price decays. Specifically, halving the distance between barriers quadruples the decay rate.
Exercise 6. The energy method requires only integration by parts and does not need the explicit form of the solution. Explain why this is an advantage over the maximum principle approach for proving uniqueness, especially for more complex PDEs with variable coefficients.
Solution to Exercise 6
The energy method proves uniqueness through three simple steps:
- Define \(E(t) = \frac{1}{2}\int u^2\,dx\)
- Show \(E'(t) \leq 0\) via integration by parts
- Conclude \(E(0) = 0 \implies E(t) = 0\)
The only analytic tool required is integration by parts. This is an advantage over the maximum principle for several reasons:
Variable coefficients: For \(\partial_t u = \frac{1}{2}\partial_x(\sigma^2(x)\partial_x u) + \mu(x)\partial_x u\), the energy method adapts naturally. Integration by parts gives \(E'(t) = -\frac{1}{2}\int \sigma^2 u_x^2\,dx + \text{drift terms}\), which can be controlled using weighted energies. The maximum principle requires uniform ellipticity and careful treatment of the drift, making it harder to apply.
Systems of equations: For coupled PDEs (e.g., multi-asset pricing), the maximum principle may fail entirely (component-wise maxima need not be controlled). Energy methods extend by defining \(E = \frac{1}{2}\int |u|^2\,dx\) as the total energy of the vector-valued solution.
Weak solutions: The energy method only requires \(u\) and \(u_x\) to be in \(L^2\) (Sobolev space \(H^1\)), not \(C^2\). The maximum principle requires classical (twice differentiable) solutions. This makes energy methods the primary tool in modern PDE theory where solutions are understood in the weak (distributional) sense.
No explicit solution needed: The proof never uses the form of the solution, only the equation itself and integration by parts.
Exercise 7. For the heat equation with a source term \(\partial_t u = \frac{1}{2}\partial_{xx}u + f(x, t)\), the energy is no longer monotonically decreasing. Compute \(E'(t)\) in this case and identify the additional term. Under what condition on \(f\) does the energy still remain bounded?
Solution to Exercise 7
For \(\partial_t u = \frac{1}{2}\partial_{xx}u + f(x,t)\) with homogeneous Dirichlet conditions:
Using integration by parts on the first term (as before):
The additional term is \(\int_a^b u\,f\,dx\), which represents the energy injected (or removed) by the source \(f\).
Bounding the energy: Apply the Cauchy-Schwarz inequality to the source term:
Using Young's inequality \(ab \leq \frac{a^2}{2\epsilon} + \frac{\epsilon b^2}{2}\) and the Poincare inequality:
For appropriate \(\epsilon\), this gives a Gronwall inequality guaranteeing \(E(t)\) remains bounded provided \(\|f(t)\|_{L^2}\) is bounded. Specifically, if \(\|f(t)\|_{L^2} \leq M\) for all \(t\), then \(E(t)\) grows at most linearly and is ultimately controlled by the balance between the dissipation rate and the source strength. In the simplest bound:
The energy remains bounded for all time whenever \(f \in L^\infty([0,T]; L^2(a,b))\).