Finite-dimensional norms and operator norms

Try

A linear map transforms every possible direction, so one sample vector cannot describe its largest amplification. For A=diag(2,−1), Euclidean output length satisfies ||Av||²=4v₁²+v₂²≤4||v||², with equality along the first axis. Its induced operator norm is therefore 2. The negative sign reverses one coordinate but does not make a length amplification negative.

Predict

Predict whether the largest matrix entry always equals the operator norm. The 2×2 all-ones matrix has largest entry one, yet it sends the unit vector (1,1)/√2 to a vector of length two. Entries depend on coordinates, while the operator norm asks for the largest output over every unit input direction. Entrywise estimates can still bound it, but dimensions and the chosen norms matter.

Experiment

Compare norms of vectors v=(x,y).

Norm	Formula	Useful comparison
Maximum	max(|x|,|y|)	At most ||v||₂
Euclidean	√(x²+y²)	At most √2||v||∞
Taxicab	|x|+|y|	At most √2||v||₂

These inequalities make the three norms equivalent in fixed dimension. They need not give the same numerical distances. Draw their unit balls only after writing the defining inequalities so the geometry does not replace the estimates.

Understand

For a linear map A between normed finite-dimensional spaces, define ||A||=sup over ||v||=1 of ||Av||. Then ||Av||≤||A||||v|| by scaling, including v=0 separately. Composition satisfies ||AB||≤||A||||B|| because each map’s bound applies successively. The norm triangle inequality follows from the pointwise triangle inequality.

Every linear map from R^n with Euclidean norm is bounded: expand Av=Σv_iAe_i, apply the triangle inequality and then Cauchy–Schwarz to obtain ||Av||≤(Σ||Ae_i||²)^(1/2)||v||₂. This finite bound also proves continuity. A matrix-entry estimate |a_ij|≤ε gives ||A||₂≤√(mn)ε via the Frobenius norm or direct coordinate inequalities; equality with the maximum entry is not asserted.

Apply

All norms on a fixed finite-dimensional real space are equivalent. Let N be any norm on R^d. The basis bound gives N(v)≤C||v||₂, which also proves |N(v)−N(w)|≤C||v−w||₂ and hence continuity. On the compact Euclidean unit sphere, N attains a positive minimum c: positivity follows because no sphere point is zero. Scaling yields c||v||₂≤N(v)≤C||v||₂.

Repair a circular proof that invokes continuity of an arbitrary norm before establishing it. The finite basis estimate supplies that step. The constants can depend on dimension and the chosen norm, so this argument does not provide dimension-free equivalence for infinite-dimensional spaces.

Continue

Optional depth observes that induced operator norms may use different norms in source and target; compatibility is essential in composition inequalities. A derivative bound written without naming norms can be ambiguous when quantitative error estimates are the goal.

Save a sharp norm calculation for a diagonal map and a nonsharp but justified bound for a general matrix. Explain whether your bound is attained and whether that matters for the application. Next, Fréchet differentiation will use operator norms to express one linear approximation that controls every sufficiently small direction, rather than isolated coordinate slopes. In finite dimensions, norm equivalence also makes convergence and differentiability independent of which equivalent norm is used. Quantitative constants change, however, so a tolerance estimate should retain its actual norm rather than relying on topological equivalence alone.