Gram-Schmidt in Inner Product Spaces
You'll learn how the Gram-Schmidt process generalizes seamlessly from Euclidean space to any abstract inner product space.
The Core Idea
You've already seen Gram-Schmidt work in ℝ² and ℝ³ using the standard dot product. The beautiful truth is: the exact same algorithm works in any inner product space—you just swap out the dot product for whatever inner product your space uses.
Whether you're working with:
- Polynomials with the integral inner product ⟨f, g⟩ = ∫₀¹ f(x)g(x) dx
- Complex vectors with ⟨u, v⟩ = u*ᴴv (conjugate transpose)
- Weighted Euclidean spaces with custom inner products
...the Gram-Schmidt recipe remains identical.
How It Works
Given linearly independent vectors {v₁, v₂, ..., vₙ} in an inner product space V:
-
Start: u₁ = v₁
-
Subtract projections: For each subsequent vector vₖ, compute:
uₖ = vₖ - ⟨vₖ, u₁⟩/⟨u₁, u₁⟩ u₁ - ⟨vₖ, u₂⟩/⟨u₂, u₂⟩ u₂ - ... - ⟨vₖ, uₖ₋₁⟩/⟨uₖ₋₁, uₖ₋₁⟩ uₖ₋₁
-
Normalize (optional): eₖ = uₖ / ‖uₖ‖
The magic: all that changes between spaces is how you compute ⟨·,·⟩ and ‖·‖. The structure of the algorithm is universal because it only relies on inner product axioms—symmetry, linearity, and positive-definiteness—which hold in every inner product space.
Why This Matters
This abstraction reveals that orthogonalization isn't tied to geometry in ℝⁿ—it's a fundamental property of any space with a notion of "perpendicular." You can orthogonalize polynomials, continuous functions, even quantum states, using the same conceptual framework.
Key Takeaway: Gram-Schmidt is not specific to Euclidean spaces; it works in any inner product space by simply using that space's inner product in place of the dot product.