Compute vector magnitude for 2D to 6D vectors using L1 (Manhattan), L2 (Euclidean), L∞ (Chebyshev), and custom Lp norms with unit vector, component contribution bars, and norm comparison.
The magnitude (or norm) of a vector measures its "size" or "length" in a vector space. While the Euclidean norm (L2) is the most familiar — it corresponds to straight-line distance — other norms like L1 (Manhattan/taxicab), L∞ (Chebyshev/max), and the generalized Lp norm each capture different notions of distance and are widely used across mathematics, data science, and engineering.
This calculator computes all major vector norms for vectors from 2D up to 6D. Enter your vector's components and see the L1, L2, L∞ norms side-by-side, along with a custom Lp norm for any p ≥ 1 that you choose. The unit vector (L2-normalized) is computed so you can extract pure direction, and the squared L2 norm is provided for optimization contexts where avoiding the square root is preferred.
A norm comparison table lists every computed norm with its formula, while a visual bar chart shows the classic inequality ‖v‖∞ ≤ ‖v‖₂ ≤ ‖v‖₁ in action. Component contribution bars break down how much each dimension contributes to the total squared magnitude, instantly revealing which axis dominates the vector's length.
Six presets span common scenarios from the Pythagorean triple (3, 4) to higher-dimensional examples. The properties table at the bottom summarizes the key axioms every norm satisfies — non-negativity, homogeneity, and the triangle inequality — plus the convergence of Lp to L∞ as p → ∞.
Understanding multiple norms is essential for machine learning (regularization uses L1 and L2), signal processing (energy is L2²), game programming (L∞ measures grid distance), and numerical analysis. This calculator makes it easy to compare norms visually and see how they diverge as vector structure changes.
Computing multiple norms (L1, L2, L∞, Lp) for the same vector requires different formulas and a custom p-norm involves exponents and roots that are easy to miscalculate. This calculator handles vectors from 2D to 6D, computes all four norms simultaneously, shows the unit vector (L2-normalized), displays component contribution bars and an inequality bar chart comparing norms, and lets you set any custom p value. It is the fastest way to compare how different norms measure the same vector and to build intuition for norm selection in optimization and machine learning.
‖v‖₁ = Σ|vᵢ|; ‖v‖₂ = √(Σvᵢ²); ‖v‖∞ = max|vᵢ|; ‖v‖_p = (Σ|vᵢ|^p)^(1/p)
Result: ‖v‖₂ = 13, ‖v‖₁ = 19, ‖v‖∞ = 12
L2: √(9 + 16 + 144) = √169 = 13. L1: |3| + |4| + |12| = 19. L∞: max(3, 4, 12) = 12. Unit vector = (3/13, 4/13, 12/13) ≈ (0.2308, 0.3077, 0.9231).
The **Lp norm** generalizes the notion of length: ‖v‖_p = (Σ|vᵢ|^p)^(1/p) for p ≥ 1. The three most common special cases are L1 (taxicab / Manhattan distance: sum of absolute values), L2 (Euclidean distance: the familiar √(Σvᵢ²)), and L∞ (Chebyshev / max distance: largest absolute component). As p increases from 1 toward ∞, the norm ball shrinks from a diamond (L1) to a circle (L2) to a square (L∞) in 2D. The inequality ‖v‖_∞ ≤ ‖v‖₂ ≤ ‖v‖₁ always holds, with equality only for vectors with a single non-zero component.
The choice of norm has practical consequences. **L2** is rotationally invariant and smooth, making it the default in physics and geometry. **L1** promotes **sparsity**: LASSO regression uses the L1 penalty to drive small coefficients to exactly zero, achieving automatic feature selection. **L∞** bounds the worst-case component, useful in control theory (stay within box constraints) and game programming (grid distance). **Lp with 1 < p < 2** balances sparsity and smoothness (elastic net). Even fractional p < 1 (not a true norm) appears in compressed sensing to encourage sparser solutions.
In **k-nearest neighbors** and clustering, the distance metric (often L2 or L1) directly affects which points are “close.” **Cosine similarity** is the dot product of L2-normalized vectors, measuring angle rather than distance. **Batch normalization** in neural networks normalizes feature vectors layer by layer. **Gradient clipping** limits ‖∇‖₂ to prevent exploding gradients. Understanding norms lets you make informed choices about regularization, distance metrics, and convergence guarantees — all of which depend on how you measure “size.”
The magnitude (or norm) measures a vector's length. The most common is the Euclidean (L2) norm: ‖v‖₂ = √(v₁² + v₂² + ⋯ + vₙ²), which gives the straight-line distance from the origin to the vector's tip.
L1 sums absolute values (taxicab distance), L2 takes the square root of summed squares (straight-line distance), and L∞ takes the maximum absolute component. They measure "size" differently: L1 ≥ L2 ≥ L∞ always holds.
Different norms capture different structural properties. L2 is rotationally invariant, L1 promotes sparse solutions, L∞ bounds the worst-case component. The choice depends on the application.
The unit vector v̂ = v / ‖v‖₂ has magnitude 1 in the L2 norm and points in the same direction as v. It extracts pure direction, discarding scale.
Yes. In data science and machine learning, vectors commonly have hundreds or thousands of dimensions. The norm formulas generalize to any number of dimensions n.
The Lp norm is a family of norms parameterized by p ≥ 1: ‖v‖_p = (Σ|vᵢ|^p)^(1/p). L1, L2, and L∞ are special cases for p = 1, 2, and ∞ respectively.