Disjoint-Composition of Norms

• Disjoint-composition of norms is a framework that decomposes an ambient norm into orthogonal components whose contributions are aggregated via simple diagonal scaling.
• It finds application in matrix analysis, Banach lattices, and convex optimization by ensuring disjointness preservation and accurate norm behavior.
• The structure enforces a block/tensor form for maps, linking triple-product identities with norm preservation conditions through unitary rotations and scaling.

“Disjoint-composition of norms” (Editor's term) denotes a structural phenomenon in which disjointness or orthogonality splits an ambient space into noninterfering components, and a linear or nonlinear map acts on those components by simple local operations while the global norm is determined by how the component contributions compose. In rectangular matrix spaces this phenomenon is explicit: disjointness-preserving maps have a rigid block/tensor form, and norm preservation reduces to simple constraints on diagonal scaling parameters; closely related decompositions also appear for disjointness preserving operators on Banach lattices and in several other norm-composition frameworks (Li et al., 2019, Schep, 2016).

1. Rectangular matrices, disjointness, and unitarily invariant norms

For a field $\mathbb{F}=\mathbb{R}$ or $\mathbb{C}$, $M_{m,n}=M_{m,n}(\mathbb{F})$ denotes the vector space of all $m\times n$ matrices over $\mathbb{F}$. For $A,B\in M_{m,n}$, with adjoint $A^*$, the relation

$$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$

defines disjointness. Geometrically, $A^*B=0_n$ means the initial spaces are orthogonal, while $AB^*=0_m$ means the ranges are orthogonal. If $\mathbb{C}$0 are partial isometries, then $\mathbb{C}$1 exactly says that their range projections and initial projections are orthogonal (Li et al., 2019).

The same matrix setting carries the Jordan $\mathbb{C}$2-triple product

$\mathbb{C}$3

which is the canonical triple product for the $\mathbb{C}$4-triple structure of rectangular matrices. It also carries the Schatten $\mathbb{C}$5-norms

$\mathbb{C}$6

and the Ky Fan $\mathbb{C}$7-norms

$\mathbb{C}$8

where $\mathbb{C}$9 are the singular values and $M_{m,n}=M_{m,n}(\mathbb{F})$0. These norms are unitarily invariant: $M_{m,n}=M_{m,n}(\mathbb{F})$1 for all unitaries of appropriate sizes (Li et al., 2019).

This matrix-theoretic setting is the primary concrete realization of disjoint-composition of norms. Disjointness is an orthogonality condition on initial and range spaces, and unitarily invariant norms depend only on singular values. A plausible implication is that any map preserving disjointness and a singular-value norm should be forced into a form that separates orthogonal components rather than mixing them.

2. Canonical block form of disjointness-preserving maps

Let $M_{m,n}=M_{m,n}(\mathbb{F})$2 be linear. The central structural theorem states that $M_{m,n}=M_{m,n}(\mathbb{F})$3 preserves disjointness if and only if there exist unitary (orthogonal) matrices $M_{m,n}=M_{m,n}(\mathbb{F})$4, $M_{m,n}=M_{m,n}(\mathbb{F})$5, and diagonal matrices $M_{m,n}=M_{m,n}(\mathbb{F})$6 with positive diagonal entries, with either $M_{m,n}=M_{m,n}(\mathbb{F})$7 or $M_{m,n}=M_{m,n}(\mathbb{F})$8 possibly empty, such that

$M_{m,n}=M_{m,n}(\mathbb{F})$9

Here $m\times n$0 is permutation-similar to a direct sum of scalar multiples of $m\times n$1, and $m\times n$2 is similarly a direct sum of scalar multiples of $m\times n$3 (Li et al., 2019).

This form has a precise geometric meaning. The map preserves disjointness by re-using $m\times n$4 and $m\times n$5 only on mutually orthogonal summands, scaled by positive factors, and then applying left and right unitaries as global rotations. There is no mixing between the “$m\times n$6-copy” and the “$m\times n$7-copy” other than via unitary equivalence. In this sense, the target space decomposes into disjoint blocks, and $m\times n$8 acts on each block by a simple scale-and-unitary rule.

Several special cases clarify the theorem. If $m\times n$9 is vacuous, then $\mathbb{F}$0 is built only from transpose copies; if $\mathbb{F}$1 is vacuous, then $\mathbb{F}$2 is built only from copies of $\mathbb{F}$3. In the square case $\mathbb{F}$4, if $\mathbb{F}$5 preserves disjointness and is nonzero, then, up to conjugation by unitaries, it has the form

$\mathbb{F}$6

This is consistent with the rigidity expected from same-size isometry and preserver results (Li et al., 2019).

3. Triple products, zero products, and normalization by partial isometries

The triple product and disjointness are linked by the equivalence

$\mathbb{F}$7

Thus zero triple products involving two equal arguments encode disjointness. It follows that any linear map preserving zero triple products necessarily preserves disjointness, so the block form above already governs zero triple-product preservers (Li et al., 2019).

More precisely, a linear map $\mathbb{F}$8 preserves zero triple products if and only if it has the same disjointness-preserving form

$\mathbb{F}$9

If $A,B\in M_{m,n}$0 is a $A,B\in M_{m,n}$1-triple homomorphism, then the diagonal weights are forced to be identities: $A,B\in M_{m,n}$2 The triple-product identity $A,B\in M_{m,n}$3 forces the scaling on each copy of $A,B\in M_{m,n}$4 to be exactly $A,B\in M_{m,n}$5. A concise way to state the outcome is that triple-product structure equals disjointness structure plus normalization (Li et al., 2019).

The same normalization appears through partial isometries. A matrix $A,B\in M_{m,n}$6 is a partial isometry iff $A,B\in M_{m,n}$7, equivalently all singular values are $A,B\in M_{m,n}$8 or $A,B\in M_{m,n}$9. The paper shows equivalences between the following conditions: $A^*$0 maps partial isometries to partial isometries; $A^*$1 sends disjoint rank-one partial isometries to disjoint partial isometries; $A^*$2 preserves disjointness and sends some nonzero partial isometry to a partial isometry; $A^*$3 preserves zero triple products and sends some nonzero partial isometry to a partial isometry; and $A^*$4 is a $A^*$5-triple homomorphism. In the language of disjoint-composition, partial isometries detect exactly when the local diagonal scaling has collapsed to identity.

4. Schatten and Ky Fan norms as norms on disjoint block sums

The matrix-space version of disjoint-composition of norms is clearest for unitarily invariant norms. If $A^*$6 and $A^*$7 are disjoint, then their singular value decompositions can be arranged to be orthogonal, so the singular values of $A^*$8 behave like concatenations of those of $A^*$9 and $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$0. For Schatten norms,

$$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$1

and for Ky Fan norms the top $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$2 singular values of $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$3 are selected from those of $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$4 and $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$5 separately. This clean behavior is exactly what makes the block form compatible with norm preservation (Li et al., 2019).

For Schatten $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$6-norms with $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$7 and $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$8, the following are equivalent: $$A\perp B \quad\Longleftrightarrow\quad A^*B=0_n,\ \ AB^*=0_m$$9 for all $A^*B=0_n$0; $A^*B=0_n$1 for all $A^*B=0_n$2 of rank at most $A^*B=0_n$3; and the existence of unitaries $A^*B=0_n$4 and positive diagonal $A^*B=0_n$5 such that

$A^*B=0_n$6

and

$A^*B=0_n$7

Norm preservation therefore reduces to a single global constraint on the diagonal scaling parameters (Li et al., 2019).

In the complex case, Ky Fan norm preservation on rank-$A^*B=0_n$8 matrices yields the same block/tensor form together with the conditions

$A^*B=0_n$9

Again the essential point is that disjoint copies do not interfere in the singular value structure, so the norm of $AB^*=0_m$0 is determined by the norm of $AB^*=0_m$1 composed with the diagonal scales $AB^*=0_m$2 (Li et al., 2019).

A concrete $AB^*=0_m$3 example makes the mechanism explicit. With

$AB^*=0_m$4

one has $AB^*=0_m$5. If $AB^*=0_m$6, $AB^*=0_m$7 is vacuous, and $AB^*=0_m$8, then $AB^*=0_m$9 and $\mathbb{C}$00 remain disjoint. For the Schatten $\mathbb{C}$01-norm,

$\mathbb{C}$02

so norm preservation requires $\mathbb{C}$03. This is the simplest illustration of how the norm of a disjoint block sum is the sum of the contributions from its blocks.

5. Banach lattices, dense ideals, and operator norms controlled by disjointness

In Banach lattices, disjointness is defined by

$\mathbb{C}$04

For a linear disjointness preserving operator $\mathbb{C}$05, the fundamental boundedness theorem states that for any disjoint sequence $\mathbb{C}$06 in $\mathbb{C}$07, there exists $\mathbb{C}$08 such that

$\mathbb{C}$09

From this one obtains an order dense ideal $\mathbb{C}$10 on which $\mathbb{C}$11 is norm bounded. If $\mathbb{C}$12 has order continuous norm, then every disjointness preserving map $\mathbb{C}$13 has a unique decomposition

$\mathbb{C}$14

where $\mathbb{C}$15 is a bounded disjointness preserving operator and $\mathbb{C}$16 is a disjointness preserving operator that is zero on a norm dense ideal. For $\mathbb{C}$17, every disjointness preserving operator is norm bounded on a norm dense sublattice algebra of $\mathbb{C}$18, and one gets a decomposition into a bounded disjointness operator and a finite sum of unbounded disjointness preserving operators concentrated at finitely many singularity points (Schep, 2016).

This gives a Banach-lattice analogue of disjoint-composition of norms. The norm behavior of a disjointness preserving operator splits into a bounded part on a large ideal and an unbounded part concentrated on a small residual set. In the $\mathbb{C}$19 case, the decomposition

$\mathbb{C}$20

localizes the unbounded behavior at finitely many singular points, so the global operator norm is assembled from a bounded regular component plus finitely many localized singular components (Schep, 2016).

A related operator-theoretic formalization appears through affiliated operator classes and enveloping norms. If $\mathbb{C}$21 is one of the disjointly defined operator classes, the enveloping norm is

$\mathbb{C}$22

equivalently

$\mathbb{C}$23

The spaces of regularly affiliated operators are Banach spaces under these enveloping norms, so domination by operators defined through disjoint convergence behavior becomes itself a norm structure (Emelyanov et al., 2022).

Disjointness can also become norm-definable. For exactly $\mathbb{C}$24-convex, strictly monotone Banach lattices with $\mathbb{C}$25,

$\mathbb{C}$26

A plausible implication is that, in such classes, the decomposition of norms along disjoint parts is already encoded in the Banach-space language, without explicit reference to lattice operations (Raynaud, 2016).

6. Other compositional frameworks and scope of the idea

The same structural motif appears in several distinct literatures. For semigroups of continuous functions, surjections $\mathbb{C}$27 satisfying a two-variable norm condition

$\mathbb{C}$28

for $\mathbb{C}$29 are forced to be composition in modulus maps: $\mathbb{C}$30 Here the preserved quantity is a composite norm built from a function $\mathbb{C}$31 of two variables, and preservation of this composite norm rigidifies the map to a boundary homeomorphism together with modulus composition (Jafarzadeh et al., 2019).

In convex analysis and learning theory, optimal interpolation norms are defined by

$\mathbb{C}$32

This infimal postcomposition framework encompasses the latent group lasso, the overlapping group lasso, and certain norms used for learning tensors, and its dual norm is

$\mathbb{C}$33

Here norm composition is explicit: block norms are aggregated by an outer norm and transported through a linear operator (Combettes et al., 2016).

Other settings exhibit the same idea in more specialized forms. $\mathbb{C}$34-norms on $\mathbb{C}$35 form a commutative semigroup under multiplication induced by componentwise multiplication of independent generators, and the idempotent $\mathbb{C}$36-norms are exactly the blockwise sup-plus-$\mathbb{C}$37 norms

$\mathbb{C}$38

determined by completely dependent blocks (Falk, 2013). For pseudo-t-norms on complete lattices, there exists a left-continuous t-norm on the ordinal sum $\mathbb{C}$39 if and only if there exists a left-continuous t-norm on $\mathbb{C}$40, and the glued operation is defined by the local t-norms on the components and meet on mixed pairs (He et al., 9 Jun 2025). In arithmetic, the multinorm principle gives a local-global statement for products of field norms from $\mathbb{C}$41 and $\mathbb{C}$42 when the Galois closures are linearly disjoint over $\mathbb{C}$43 (Pollio et al., 2012).

A broader noncommutative analogue appears for maps $\mathbb{C}$44. The norms

$\mathbb{C}$45

define matrix-level control by order intervals, and a map is completely order bounded when the ampliations $\mathbb{C}$46 are uniformly bounded with respect to these norms. In the cases $\mathbb{C}$47 or $\mathbb{C}$48, with $\mathbb{C}$49 injective, completely order bounded maps are decomposable; for $\mathbb{C}$50, there are uniformly bounded ampliations that are not decomposable (Neuhardt, 2019).

These examples do not constitute a single standard theory. This suggests that “disjoint-composition of norms” names a recurring structural motif: disjointness, orthogonality, or linearly disjoint decomposition suppresses interference between components, and once that suppression is present, norm preservation typically reduces to a small collection of global parameters, algebraic constraints, or localization rules.