How is it related to Laplace's equation?

Laplace's equation is the zero-source case, ∇²u = 0.

Why do boundary conditions matter?

They constrain the unknown field and help determine whether a solution exists and whether it is unique.

Does every Poisson problem have one unique solution?

No. Existence and uniqueness depend on the domain, source and boundary conditions; pure Neumann problems require particular care.

AI Glossary / Computer vision and multimodal AI / Intermediate

Poisson equation

Definition

The Poisson equation is a partial differential equation that relates the Laplacian of an unknown function to a specified source term. Commonly written as ∇²u = f, it appears in physical modeling and image reconstruction, with boundary conditions helping determine the solution.

What it means

The unknown u might represent a physical potential, a steady temperature field or image intensities. Its Laplacian combines second derivatives and describes how the function differs locally from surrounding values. The right-hand side f supplies a specified pattern of forcing or curvature. When f is zero, the equation becomes Laplace's equation. The name Poisson also appears in a probability distribution, but solving this equation is not the same as modeling Poisson noise. The meaning of each quantity comes from the application and the chosen sign convention.

How it works

Define a spatial domain, the source term and boundary conditions. Dirichlet conditions specify values on the boundary; Neumann conditions specify normal derivatives, which can represent a flux. A numerical method approximates derivatives and solves the resulting system for values inside the domain.
On a unit-spaced two-dimensional grid, a common discrete Laplacian is the sum of four direct neighbors minus four times the center value. Other spacing or discretization choices change the coefficients. Appropriate Dirichlet conditions often give a unique solution, while a pure Neumann problem requires a compatibility condition and generally determines the solution only up to an additive constant. Boundary information is therefore essential, not an optional finishing step.

Three practical examples

A gradient-domain image editor reconstructs intensities from a guidance field and destination boundary values.
An engineering simulation models a steady field with a distributed source.
A numerical-methods class compares iterative solvers for a sparse system produced by a grid-based Poisson problem.

Think of it this way

Think of a flexible surface whose local bending is prescribed while its edges are constrained. The interior must satisfy both the local requirements and the boundary information. This is only a conceptual guide: the exact physical interpretation depends on the model, and a mathematical source term need not represent a literal force on a physical sheet.

Worked example

Take one unknown grid value u with four known direct neighbors: 10, 20, 30 and 40. With unit spacing and the convention ∇²u = neighbor sum − 4u, a zero source gives 100 − 4u = 0, so u = 25. This is the local averaging property associated with the discrete Laplace equation.
If the source term is 20, the equation becomes 100 − 4u = 20 and u = 20. A different sign convention would change how the source is written. In a real multi-pixel problem, the neighboring unknowns are coupled, so solving one point independently is generally insufficient.

Illustrative case study

In a fictional photo-editing prototype, a team reconstructs a selected region from desired image gradients. An early implementation produces an unexpected overall brightness shift because the boundary handling is incorrect. The team tests a small grid with known boundary values and checks the residual of the discrete equations before returning to full images. It then evaluates visual effects separately from solver correctness. The case demonstrates why a mathematically accurate interior calculation can still produce the wrong application result when boundary conditions or conventions are mismatched.

Common misconceptions and mistakes

The Poisson equation is not the Poisson probability distribution. Its source term need not represent random noise. A solver also does not determine a meaningful answer from the right-hand side alone without appropriate boundary or reference information. Changing grid spacing without changing derivative coefficients alters the problem. Another mistake is judging convergence only by a visually smooth image: a result can look smooth while failing to satisfy the specified equations or boundary values accurately.

Business use

For software using Poisson reconstruction, separate the application model from the numerical solver. Document the domain, sign convention, spacing, source construction and boundary treatment. Check simple known solutions and monitor a suitable residual or convergence criterion. For large images or meshes, memory use and solver speed can matter, but faster convergence does not compensate for solving the wrong model. Review whether the final output fulfills the visual or physical purpose as well as the equations.

Check your understanding

Under the discrete convention neighbor sum − 4u = 0, what is the center value when the four neighbors are 8, 12, 16 and 20?

Answer: Their sum is 56, so u = 56/4 = 14. This local result assumes unit spacing and the stated zero-source equation.

Related terms

Poisson (gradient-domain) blending
Laplacian
partial differential equation
Laplace equation
boundary condition
Dirichlet condition
Neumann condition
finite difference
sparse linear system
Poisson blending

Learn next

Learn the discrete Laplacian and boundary conditions next. Solve a tiny grid by hand, then distinguish the equation residual from whether the model represents the intended image-editing or physical problem.

Sources

MIT OpenCourseWare, introduction to Laplace and Poisson equations: https://ocw.mit.edu/courses/18-152-introduction-to-partial-differential-equations-fall-2011/4336b9d0607ac33f430908fdf4cb78ba_MIT18_152F11_lec_06.pdf MIT OpenCourseWare, applied mathematics discussion: https://ocw.mit.edu/courses/18-086-mathematical-methods-for-engineers-ii-spring-2006/d9ad8a20efdcd46c4cccd981bc4306e5_am37.pdf Primary teaching-source search excerpts consulted; full source and editorial review remain pending.

Previous: Poisson distribution Next: Poisson (gradient-domain) blending