Visualizing the Local Linear Approximation
The local linear approximation of a scalar-valued function has a beautiful geometric interpretation: it is the tangent plane to the surface at the point of approximation.
The tangent plane at has the equation . This plane touches the surface at exactly the point , and near this point the surface is well approximated by the plane.
The tangent plane diagram helps build intuition: moving in the -direction, the plane rises with slope ; moving in the -direction, it rises with slope . The LLA says the actual surface has the same slope in both coordinate directions — and in all other directions as well.
Formal View
This plane is the unique plane through that has the same partial derivatives in and as at .
Interactive Visualization
Local Linear Approximation
Why This Matters
Tangent planes are the multivariable analog of tangent lines and underlie virtually all first-order optimization.
- Gradient descent follows the tangent plane approximation to step toward a minimum
- Computer graphics: normal vectors to surfaces are perpendicular to tangent planes
- Numerical methods: Newton's method uses tangent hyperplanes to solve systems of equations
Quiz
The tangent plane to at is the graph of which function?
At a local minimum of , the tangent plane is horizontal (parallel to the -plane).
Common Mistakes
- Confusing the tangent plane equation with the surface itself — the plane equals the surface only at the point of tangency.
- Forgetting the constant term when writing the tangent plane equation.
- Using the tangent plane too far from the point of tangency where the approximation error becomes large.