Math/Projection

• Projection

  • Overview

    • A projection is a way of turning something into the part of itself that matters for a chosen direction, subspace, coordinate system, or question.
    • In vector geometry, the central picture is shadow-casting: a vector can be "dropped" onto a line or plane, producing the component that lies along that target and leaving behind a remainder that points away from it.
    • The point is not merely to make a smaller object. A projection separates a thing into seen-through-this-lens and left-over-after-that-lens.

  • Context

    • Projection appears whenever mathematics asks: what part of this object belongs to this simpler structure?
    • A 3D point projected onto a 2D screen keeps the information needed for the screen and discards depth.
    • A vector projected onto a direction keeps the amount of motion in that direction and discards the perpendicular motion.
    • A dataset projected onto a smaller set of coordinates keeps the variation expressible in those coordinates and treats the rest as residual.

  • Vector Projection

    • Given a vector v and a nonzero vector u, the projection of v onto the line spanned by u is the part of v that points in the same direction as u.
    • The usual formula is:
      • proj_u(v) = ((v . u) / (u . u)) u
    • The dot product v . u measures how much v aligns with u.
    • Dividing by u . u normalizes for the length of u, so the result depends on the direction of u, not on how long the chosen direction vector happens to be.
    • If u is a unit vector, the formula becomes simpler:
      • proj_u(v) = (v . u) u
    • The scalar v . u is the signed length of the shadow of v along u: positive if it points with u, negative if it points against u, and zero if it is perpendicular.

  • Mechanism

    • Projection splits a vector into two pieces:
      • v = projected part + residual part
    • For orthogonal projection, the residual is perpendicular to the target line or subspace.
    • This perpendicular residual is what makes orthogonal projection feel like the "closest point" on the target: among all vectors on the target line or subspace, the projected vector is the one nearest to the original.
    • The same idea generalizes from projecting onto a line to projecting onto a plane or any subspace.

  • Key Principles

    • A projection chooses a target. Nothing is projected "in general"; it is projected onto a line, plane, subspace, basis, screen, or model.
    • A projection keeps what the target can express. The projected object is the best version of the original that lives inside the chosen target.
    • A projection creates a residual. What does not fit the target has not vanished conceptually; it becomes the error, remainder, perpendicular component, or lost information.
    • Orthogonal projection is closest-fit projection. In Euclidean geometry, dropping at a right angle gives the nearest point on the target.
    • Projection can be lossy or exact. Projecting onto a lower-dimensional target usually loses information, while projecting an object already in the target changes nothing.

  • Idempotence

    • Many mathematical projections have the property P(P(x)) = P(x).
    • Once an object has been projected onto the target, projecting it again does not change it.
    • This is why projections often behave like filters: after the off-target part has been removed, the operation has nothing more to remove.

  • Examples

    • A shadow on the ground is a physical projection from 3D space onto a 2D plane.
    • Resolving a force into horizontal and vertical components projects the force vector onto coordinate axes.
    • Least-squares regression projects an observed vector of data onto the subspace spanned by the model's predictors; the residual is the part the model cannot explain.
    • Computer graphics projects 3D geometry onto a 2D image plane so it can be drawn on a screen.
    • A conceptual model can act like a projection when it preserves some structure of a situation and deliberately suppresses other structure.

  • Misconceptions

    • Projection is not the same thing as scaling a vector down. Scaling keeps the vector on its original line; projection changes the vector so that it lies in the target.
    • Projection is not always visual perspective. Perspective projection in graphics is one kind; orthogonal projection in linear algebra is another.
    • The residual is not necessarily "mistake" in a moral sense. It is the part of the original that the chosen target cannot represent.
    • A projection is only meaningful relative to the choice of target and geometry. Change the target, distance measure, or inner product, and the projected result can change.

  • Why It Matters

    • Projection is one of mathematics' basic tools for turning complex structure into tractable structure.
    • It lets a problem be decomposed: part along this direction, part outside it; part explained by this model, part not explained; part visible in this representation, part hidden.
    • For vectors, projection is the bridge between geometry and algebra: the intuitive shadow becomes a dot-product formula, and the formula becomes a reusable operation across higher-dimensional spaces.