Canonical: https://knowlpedia.com/functional-analysis/graph-norm/

# Graph norm

The norm on an operator domain that controls both a vector and its image.

Let \(T:D(T)\subseteq X\to Y\) be a linear operator between [normed vector spaces](https://knowlpedia.com/linear-algebra/normed-vector-space/). The **graph norm** of \(T\) is the norm on \(D(T)\) defined by
\[
\lVert x\rVert_T=\lVert x\rVert_X+\lVert Tx\rVert_Y.
\]
Equivalently, it is the norm transported from the [graph of \(T\)](https://knowlpedia.com/functional-analysis/graph-of-operator/) by the linear bijection \(x\mapsto(x,Tx)\), when \(X\times Y\) is given the sum norm. The graph norm is stronger than the norm inherited from \(X\): convergence \(x_n\to x\) in graph norm means both \(x_n\to x\) in \(X\) and \(Tx_n\to Tx\) in \(Y\).

## Completeness and closedness

If \(X\) and \(Y\) are [Banach spaces](https://knowlpedia.com/linear-algebra/banach-space/), then \(T\) is closed exactly when \((D(T),\lVert\cdot\rVert_T)\) is a Banach space. Indeed, the graph map identifies the operator domain isometrically with \(\Gamma(T)\subseteq X\times Y\), and a [linear subspace](https://knowlpedia.com/convex-analysis/linear-subspace/) of a Banach space is complete precisely when it is closed.

## Equivalent choices

When \(X\) and \(Y\) are [Hilbert spaces](https://knowlpedia.com/linear-algebra/hilbert-space/), one often uses
\[
\lVert x\rVert_{T,2}
=\bigl(\lVert x\rVert_X^2+\lVert Tx\rVert_Y^2\bigr)^{1/2}.
\]
This norm is equivalent to the sum graph norm and comes from the graph [inner product](https://knowlpedia.com/linear-algebra/inner-product/)
\[
\langle x,z\rangle_T=\langle x,z\rangle_X+\langle Tx,Tz\rangle_Y.
\]
The two formulas define the same topology on \(D(T)\), though only the second is induced by an inner product.

## Cores

For a [closed operator](https://knowlpedia.com/functional-analysis/closed-linear-operator/) \(T\), a subspace \(D_0\subseteq D(T)\) is a **core** when it is dense in \(D(T)\) for the graph norm. Equivalently, the restriction \(T|_{D_0}\) is closable and its closure is \(T\). Ordinary density of \(D_0\) in \(X\) is not enough, because it does not control convergence of the images \(Tx\).

## References

1. Tosio Kato, *Perturbation Theory for Linear Operators*, Springer, 1995 reprint of the 2nd ed. [Springer DOI record](https://doi.org/10.1007/978-3-642-66282-9). Relevant: Chapter III.
2. Konrad Schmüdgen, *Unbounded Self-adjoint Operators on Hilbert Space*, Springer, 2012. [Springer DOI record](https://doi.org/10.1007/978-94-007-4753-1). Relevant: Chapter 1.
