Skip to content

Nuclear Norm (Trace Norm)

Formula

\[ \|A\|_* = \sum_{i=1}^{\min(m,n)} \sigma_i(A) \]

Parameters

  • \(A\in\mathbb{R}^{m\times n}\): matrix
  • \(\sigma_i(A)\): singular values of \(A\) (nonnegative)

What it means

The nuclear norm measures matrix size by summing its singular values.

What it's used for

  • Convex surrogate for matrix rank in optimization.
  • Low-rank matrix completion and denoising.

Key properties

  • \(\|A\|_* = \operatorname{tr}(\Sigma)\) when \(A = U\Sigma V^T\) (SVD).
  • Dual norm of the spectral norm (\(\|\cdot\|_2\) for matrices).
  • Convex, unlike rank.

Common gotchas

  • Do not confuse with Frobenius norm (\(\sqrt{\sum_i \sigma_i^2}\)).
  • “Trace norm” equals nuclear norm even for non-square matrices, but it is not generally \(\operatorname{tr}(A)\).

Example

If \(A\) has singular values \(3,1,0\), then \(\|A\|_* = 4\).

How to Compute (Pseudocode)

Input: matrix A
Output: nuclear norm ||A||_*

compute singular values sigma_i of A (for example, via SVD)
return sum of singular values

Complexity

  • Time: Dominated by singular-value computation (dense SVD/eigensolver workflow)
  • Space: Depends on whether full SVD factors or only singular values are computed
  • Assumptions: Exact computation shown; large-scale applications often use partial/approximate spectral methods

See also