Classification: Elliptic, Parabolic, Hyperbolic

Learning Objectives

By the end of this section, you will be able to:

1. Compute the discriminant $B^2 - 4AC$ for any second-order linear PDE and determine its classification
2. Explain the geometric connection between PDE classification and conic sections (ellipse, parabola, hyperbola)
3. Describe the characteristic curves for each type and what they reveal about information propagation
4. Compare the solution behavior of elliptic, parabolic, and hyperbolic PDEs using physical intuition
5. Determine the appropriate initial and boundary conditions for each PDE type (well-posedness)
6. Recognize variable-coefficient and mixed-type PDEs that change classification across the domain
7. Connect PDE classification to modern machine learning applications including diffusion models and neural operators

The Big Picture: Why Classification Matters

"In mathematics, you don't understand things. You just get used to them." — John von Neumann. But PDE classification is one place where understanding truly helps: knowing the type tells you everything about how solutions behave.

Imagine you are an engineer confronted with a partial differential equation describing some physical system. Before you can solve it, choose a numerical method, or even decide what boundary data to provide, you need to answer one fundamental question: What type of PDE is this?

The classification of second-order PDEs into elliptic, parabolic, and hyperbolic types is not merely a mathematical taxonomy. It determines:

The remarkable fact is that a single number — the discriminant $B^2 - 4AC$ — captures all of this information. This is one of the most elegant results in applied mathematics.

The General Second-Order Linear PDE

The most general second-order linear PDE in two independent variables $x$ and $y$ has the form:

The classification depends only on the highest-order terms — the coefficients $A$, $B$, and $C$ of the second-order derivatives. The lower-order terms (involving $D, E, F, G$) affect the specific solution but not the fundamental type of the equation.

The Principal Part

The principal part (or principal symbol) of the PDE consists of only the highest-order terms:

This is the part that determines the character of the PDE. Think of it as the "skeleton" of the equation — the lower-order terms add flesh, but the skeleton determines the shape.

The Discriminant Condition

The classification of a second-order PDE mirrors the classification of quadratic equations. Recall that for the quadratic $ax^2 + bx + c = 0$, the discriminant $b^2 - 4ac$ determines the nature of the roots: two real roots, one repeated root, or two complex roots.

For PDEs, we form the characteristic equation by replacing partial derivatives with direction variables. The PDE $Au_{xx} + Bu_{xy} + Cu_{yy} = 0$ leads to the characteristic quadratic:

The discriminant of this quadratic is:

Worked Examples

The Conic Section Connection

The names "elliptic," "parabolic," and "hyperbolic" come directly from conic sections, and this is not a coincidence. Consider the quadratic form:

The discriminant $B^2 - 4AC$ classifies this curve as:

The connection goes deeper than naming. The principal symbol of the PDE, $A\xi^2 + B\xi\eta + C\eta^2$, defines a quadratic form in the frequency/momentum variables $(\xi, \eta)$. The level sets of this quadratic form are exactly these conic sections. The geometry of these curves encodes how information propagates through the solution.

Elliptic PDEs: Equilibrium and Steady State

Elliptic PDEs describe steady-state or equilibrium phenomena — situations where a system has settled into a configuration that balances all internal forces.

The Prototype: Laplace's Equation

This equation says: the value of u at every point equals the average of its neighbors. There are no local maxima or minima in the interior of the domain — all extrema occur on the boundary. This is the maximum principle, the defining property of elliptic PDEs.

Physical Meaning

Think of stretching a rubber membrane across a wire frame of arbitrary shape. The membrane settles into a shape where every point is at the average height of its neighbors. This equilibrium shape satisfies Laplace's equation.

Key Properties of Elliptic PDEs

1. No time variable: Elliptic PDEs describe static configurations, not evolution. There is no "before" or "after."
2. Maximum principle: The solution attains its maximum and minimum values on the boundary, never in the interior.
3. Infinite smoothness: Solutions to elliptic PDEs are $C^{\infty}$ (infinitely differentiable) in the interior, even if the boundary data is rough.
4. Global dependence: The solution at every interior point depends on the entire boundary. Changing the boundary anywhere affects the solution everywhere.
5. Boundary value problems: Well-posed with Dirichlet, Neumann, or Robin conditions on the boundary. No initial conditions needed.

Important Elliptic PDEs

Parabolic PDEs: Diffusion and Smoothing

Parabolic PDEs describe diffusive processes — phenomena where sharp features smooth out over time and the system evolves toward equilibrium.

The Prototype: The Heat Equation

This equation is first-order in time and second-order in space. The mismatch between temporal and spatial orders is the hallmark of parabolic PDEs. It gives them their distinctive one-way-in-time character.

Physical Meaning

Drop a spoonful of dye into a glass of water. The dye spreads out, its concentration evening out over time. Sharp edges blur. This irreversible smoothing process is governed by the heat equation. You cannot "un-mix" the dye — parabolic equations are irreversible.

Key Properties of Parabolic PDEs

1. Irreversibility: Parabolic PDEs run in only one time direction. Running the heat equation backward is ill-posed — small errors grow exponentially.
2. Infinite speed of propagation: A localized disturbance is felt everywhere instantly, though the effect decays rapidly with distance.
3. Smoothing: Solutions immediately become smooth ( $C^{\infty}$), even if the initial data has discontinuities.
4. Exponential decay of modes: High-frequency components decay as $e^{-n^2\pi^2\alpha t}$. Higher frequencies decay much faster, explaining the smoothing.
5. One initial condition: Only $u(x, 0) = f(x)$ is needed (the state at $t = 0$), plus boundary conditions.

Important Parabolic PDEs

Hyperbolic PDEs: Waves and Propagation

Hyperbolic PDEs describe wave propagation — phenomena where disturbances travel at finite speed without dissipation.

The Prototype: The Wave Equation

This equation is second-order in both time and space. The symmetry between temporal and spatial orders gives hyperbolic equations their wave-like, reversible character.

Physical Meaning

Pluck a guitar string. The resulting vibration travels back and forth along the string at speed $c$, reflecting off the fixed ends. The wave shape is preserved — it doesn't spread out or smooth like heat. If you could record the motion and play it backward, it would look equally physical. This is the reversibility of hyperbolic equations.

Key Properties of Hyperbolic PDEs

1. Finite speed of propagation: Disturbances travel at speed $c$, defining a light cone or domain of dependence. Points outside this cone are unaffected.
2. Two initial conditions: Both $u(x, 0) = f(x)$ (displacement) and $u_t(x, 0) = g(x)$ (velocity) are needed.
3. Sharp features preserved: Discontinuities and sharp fronts travel along characteristics without being smoothed. Shocks can form.
4. Energy conservation: The total energy $E = \frac{1}{2}\int (u_t^2 + c^2 u_x^2)\,dx$ is preserved over time. No dissipation occurs.
5. Reversibility: The equation is symmetric under $t \to -t$. You can run the equation backward.

Important Hyperbolic PDEs

Characteristic Curves: The DNA of PDEs

Characteristic curves are the most powerful concept in PDE theory. They are special curves in the $(x, y)$ or $(x, t)$ plane along which the PDE reduces to an ordinary differential equation.

Mathematical Definition

For the PDE $Au_{xx} + Bu_{xy} + Cu_{yy} = 0$, the characteristic curves satisfy:

Solving this quadratic gives the slopes of the characteristics:

What Characteristics Tell Us

Characteristics are not just mathematical abstractions. They have profound physical meaning:

• Elliptic (no real characteristics): There are no special directions. Information spreads isotropically through the domain. The solution at any point depends on the entire boundary.
• Parabolic (one family): There is one special direction (typically, lines of constant time). Information diffuses outward from initial data.
• Hyperbolic (two families): There are two special directions along which signals propagate. For the wave equation, these are $x - ct = \text{const}$ and $x + ct = \text{const}$ — the paths of right-going and left-going waves.

Domain of Dependence and Range of Influence

For hyperbolic PDEs, the characteristics define two crucial regions:

Solution Behavior Comparison

The three PDE types produce fundamentally different solution behaviors. The animation below shows solutions of each type starting from the same initial shape (a combination of sine waves):

Observe the key differences:

Well-Posedness: Getting the Conditions Right

A PDE problem is well-posed (in the sense of Hadamard) if it satisfies three conditions:

1. Existence: A solution exists.
2. Uniqueness: The solution is unique.
3. Continuous dependence: Small changes in the data produce small changes in the solution.

Each PDE type requires specific types of conditions to be well-posed. Providing the wrong type of conditions leads to ill-posed problems — solutions that don't exist, aren't unique, or are wildly sensitive to perturbations.

Variable Coefficients and Mixed Type

When the coefficients $A, B, C$ depend on position $(x, y)$, the discriminant $B^2 - 4AC$ can change sign across the domain. The PDE then changes type from region to region.

The Tricomi Equation

The most famous mixed-type PDE is the Tricomi equation:

Here $A = y, B = 0, C = 1$, so $D = -4y$:

The Tricomi equation models transonic flow — the transition between subsonic (elliptic) and supersonic (hyperbolic) flight. The line $y = 0$ represents the sonic barrier, where the flow speed equals the speed of sound.

Machine Learning Connections

PDE classification plays a surprisingly important role in modern machine learning. Understanding the type of PDE involved helps design better algorithms.

1. Diffusion Models = Parabolic PDEs

Score-based diffusion models (DALL-E, Stable Diffusion, Midjourney) are fundamentally parabolic. The forward process adds noise governed by:

This is a Fokker-Planck equation — a parabolic PDE. The smoothing property of parabolic equations (destroying information) is exactly what converts images to noise. The reverse process (generating images) inverts this diffusion by learning the score function $\nabla_x \log p(x,t)$.

Understanding that this is parabolic tells us:

• The forward process is irreversible — information is truly lost, which is why we need a neural network to learn the reverse.
• Higher frequencies are destroyed first (smoothing), so coarse structure is generated before fine details during sampling.
• The process naturally converges to a Gaussian (the steady-state equilibrium of the heat equation).

2. Neural Operators and PDE Type

Neural operators like the Fourier Neural Operator (FNO) learn to solve families of PDEs. Their architecture is designed differently depending on the PDE type:

• Elliptic PDEs: The operator needs global receptive fields because changing boundary data affects the entire solution. Fourier layers work well.
• Hyperbolic PDEs: Locality matters because of finite propagation speed. Attention mechanisms with position encoding capture the characteristic structure better.
• Parabolic PDEs: Time-stepping architectures (autoregressive) are natural because the equation is first-order in time.

3. Graph Neural Networks as Discrete Heat Equations

Message-passing in GNNs is a discrete parabolic PDE. The update rule:

is exactly a forward Euler discretization of the heat equation on a graph. This explains over-smoothing in deep GNNs: too many layers correspond to running the heat equation for too long, causing all node features to converge to the same value (the steady-state equilibrium).

4. Physics-Informed Neural Networks (PINNs)

When training PINNs, the PDE type affects:

• Loss function design: Elliptic PDEs need boundary loss terms; parabolic need initial + boundary; hyperbolic need two initial conditions.
• Training difficulty: Hyperbolic PDEs are hardest to train because sharp features and characteristics make the loss landscape rugged.
• Sampling strategy: Collocation points should respect the domain of dependence for hyperbolic equations.

Python: Automatic PDE Classifier

Let's build a Python tool that classifies any second-order PDE given its coefficients and computes the characteristic slopes:

1import numpy as np
2from enum import Enum
3
4class PDEType(Enum):
5    ELLIPTIC = "Elliptic"
6    PARABOLIC = "Parabolic"
7    HYPERBOLIC = "Hyperbolic"
8
9def classify_pde(A: float, B: float, C: float) -> tuple[PDEType, float]:
10    """
11    Classify a second-order linear PDE of the form:
12    A*u_xx + B*u_xy + C*u_yy + (lower order terms) = 0
13
14    Returns (type, discriminant).
15    The discriminant D = B^2 - 4AC determines:
16      D < 0  ->  Elliptic   (e.g., Laplace, Poisson)
17      D = 0  ->  Parabolic  (e.g., Heat/Diffusion)
18      D > 0  ->  Hyperbolic (e.g., Wave)
19    """
20    D = B**2 - 4 * A * C
21
22    if D < -1e-10:
23        return PDEType.ELLIPTIC, D
24    elif D > 1e-10:
25        return PDEType.HYPERBOLIC, D
26    else:
27        return PDEType.PARABOLIC, D
28
29def characteristic_slopes(A: float, B: float, C: float):
30    """
31    Compute the characteristic slopes dy/dx for the PDE.
32
33    The characteristic equation is:
34    A*(dy)^2 - B*(dy)(dx) + C*(dx)^2 = 0
35
36    which gives: dy/dx = (B +/- sqrt(B^2 - 4AC)) / (2A)
37    """
38    D = B**2 - 4 * A * C
39
40    if abs(A) < 1e-10:
41        return None  # Degenerate case
42
43    if D < -1e-10:
44        # Elliptic: complex slopes
45        real_part = B / (2 * A)
46        imag_part = np.sqrt(-D) / (2 * A)
47        return (complex(real_part, imag_part),
48                complex(real_part, -imag_part))
49    elif D > 1e-10:
50        # Hyperbolic: two real slopes
51        sqrt_D = np.sqrt(D)
52        return ((B + sqrt_D) / (2 * A),
53                (B - sqrt_D) / (2 * A))
54    else:
55        # Parabolic: one repeated slope
56        slope = B / (2 * A)
57        return (slope, slope)
58
59# Example classifications
60examples = [
61    ("Laplace equation", 1, 0, 1),
62    ("Heat equation (x,t)", 1, 0, 0),
63    ("Wave equation (x,t)", 1, 0, -1),
64    ("Mixed: 2u_xx + 3u_xy + u_yy", 2, 3, 1),
65    ("Tricomi at y=1: u_xx + u_yy", 1, 0, 1),
66    ("Tricomi at y=-1: -u_xx + u_yy", -1, 0, 1),
67]
68
69print("PDE Classification Results")
70print("=" * 60)
71
72for name, A, B, C in examples:
73    pde_type, D = classify_pde(A, B, C)
74    slopes = characteristic_slopes(A, B, C)
75
76    print(f"\n{name}")
77    print(f"  A={A}, B={B}, C={C}")
78    print(f"  D = B^2 - 4AC = {D:.2f}")
79    print(f"  Type: {pde_type.value}")
80    if slopes:
81        print(f"  Char. slopes: {slopes[0]:.4f}, {slopes[1]:.4f}")

Now let's see how to compute the canonical form transformation:

1import numpy as np
2
3def canonical_transform(A, B, C):
4    """
5    Find the change of variables that transforms
6    A*u_xx + B*u_xy + C*u_yy = 0
7    into its canonical form.
8
9    Returns (type, canonical_form_description, transformation)
10    """
11    D = B**2 - 4 * A * C
12
13    if D < -1e-10:
14        # ELLIPTIC -> canonical form: u_ss + u_nn = 0
15        # Use rotation to eliminate cross term
16        if abs(B) < 1e-10:
17            theta = 0
18        else:
19            theta = 0.5 * np.arctan2(B, A - C)
20
21        return {
22            "type": "Elliptic",
23            "canonical": "u_ss + u_nn = 0  (Laplace form)",
24            "rotation_angle": np.degrees(theta),
25            "transform": f"Rotate by {np.degrees(theta):.1f} degrees",
26            "meaning": "Both second derivatives have same sign"
27        }
28
29    elif D > 1e-10:
30        # HYPERBOLIC -> canonical form: u_ss - u_nn = 0
31        # or equivalently: u_ab = 0 (using characteristic coords)
32        sqrt_D = np.sqrt(D)
33        slope1 = (B + sqrt_D) / (2 * A)
34        slope2 = (B - sqrt_D) / (2 * A)
35
36        return {
37            "type": "Hyperbolic",
38            "canonical": "u_ab = 0  (d'Alembert form)",
39            "alt_canonical": "u_ss - u_nn = 0  (wave form)",
40            "char_slopes": (slope1, slope2),
41            "transform": f"a = y - {slope1:.3f}x, b = y - {slope2:.3f}x",
42            "meaning": "Second derivatives have opposite signs"
43        }
44
45    else:
46        # PARABOLIC -> canonical form: u_nn = 0 or u_t = u_xx
47        slope = B / (2 * A) if abs(A) > 1e-10 else 0
48
49        return {
50            "type": "Parabolic",
51            "canonical": "u_ss = 0  (or u_t = u_nn with lower terms)",
52            "char_slope": slope,
53            "transform": f"s = y - {slope:.3f}x,  n = x",
54            "meaning": "One second derivative vanishes"
55        }
56
57# Demonstrate canonical forms
58print("Canonical Form Transformations")
59print("=" * 50)
60
61test_cases = [
62    ("Laplace", 1, 0, 1),
63    ("Rotated Laplace", 1, 1, 1),
64    ("Wave", 1, 0, -1),
65    ("General hyperbolic", 2, 5, 1),
66    ("Heat-like", 1, 2, 1),
67]
68
69for name, A, B, C in test_cases:
70    result = canonical_transform(A, B, C)
71    print(f"\n{name}: {A}u_xx + {B}u_xy + {C}u_yy = 0")
72    print(f"  Type: {result['type']}")
73    print(f"  Canonical: {result['canonical']}")
74    print(f"  Transform: {result['transform']}")
75    print(f"  Meaning: {result['meaning']}")

Common Pitfalls

Test Your Understanding

Summary

The classification of second-order PDEs into elliptic, parabolic, and hyperbolic types is one of the most powerful organizational principles in applied mathematics. A single number — the discriminant — reveals the fundamental character of the equation.

The Classification at a Glance

Key Takeaways

1. The discriminant $D = B^2 - 4AC$ of the principal part determines the PDE type. This is computed from only the highest-order coefficients.
2. The classification mirrors conic sections: the level sets of the principal symbol $A\xi^2 + B\xi\eta + C\eta^2$ are ellipses, parabolas, or hyperbolas.
3. Elliptic PDEs describe equilibrium with no time evolution, global dependence on boundary data, and maximum principles.
4. Parabolic PDEs describe irreversible diffusion, with instantaneous smoothing and exponential decay of high-frequency modes.
5. Hyperbolic PDEs describe wave propagation at finite speed, with characteristic curves defining domains of dependence and influence.
6. Well-posedness requires matching the right type of initial and boundary conditions to the PDE type. Wrong conditions lead to ill-posed problems.
7. Variable-coefficient PDEs can change type across the domain, with important physical applications like transonic flow.
8. PDE classification directly informs ML architecture design: diffusion models are parabolic, GNN over-smoothing is a heat equation phenomenon, and neural operators must respect the propagation character of the underlying PDE.

Coming Next: In the next section, we'll explore Boundary and Initial Conditions in detail. You'll learn the precise mathematical formulations of Dirichlet, Neumann, and Robin conditions, and understand how to set up well-posed PDE problems for both analytical and numerical solution.