Physics-Preserving Linear Mapping

• Physics-preserving linear mapping is a concept where a linear transformation retains key physical invariants such as orthogonality, trace norm, or symplectic form.
• The mapping constraints force linear maps to collapse into highly structured forms, including similarities, real-linear or complex-linear isometries, and unitary or antiunitary operators.
• Extensive research illustrates applications in quantum mechanics, Hamiltonian beam dynamics, and numerical interpolation, addressing both exact and approximate invariant preservation.

“Physics-preserving linear mapping” denotes a family of transformation problems in which the admissible map is constrained not primarily by algebraic linearity alone, but by preservation of a physically or geometrically meaningful structure. In the cited literature, the preserved structure varies by setting: Hilbert-space orthogonality and angle, operator-algebraic covariance under Moore–Penrose inversion, Lorentz-cone spectral data, trace-norm distinguishability of quantum states, separability of multipartite pure states, tensor-product norms and spectra, canonical symplectic form in Hamiltonian beam dynamics, and positivity or data boundedness in numerical field transfer. The cited works suggest that the subject is best understood as a rigidity theory: once a map preserves the relevant invariant exactly, or even approximately under suitable hypotheses, its form often collapses to similarities, real-linear or complex-linear isometries, unitary or antiunitary implementations, orthogonal block conjugations, or symplectic maps (Li et al., 20 Mar 2025, Busch, 2013, Abell et al., 2022).

1. Conceptual scope and preserved structures

The common feature across this literature is the choice of an invariant regarded as structurally indispensable in the underlying model. In inner-product geometry, that invariant is often orthogonality or a fixed angle. In quantum-state spaces it is trace norm, orthogonality of positive operators, or the extremal set of pure or separable pure states. In operator algebras it can be covariance of the Moore–Penrose inverse under conjugation by invertibles. In cone geometry it is the Lorentz spectrum defined by an eigenvalue complementarity problem. In accelerator physics it is the canonical symplectic condition $M^TJM=J$. In structured-mesh interpolation it is positivity or intervalwise data boundedness rather than linearity of the transfer rule itself (Moslehian et al., 2016, Alizadeh, 2018, Bueno et al., 2022, Ouermi et al., 2023).

A second organizing distinction is between exact and approximate preservation. Exact orthogonality preservation, exact covariance-set preservation, or exact trace-norm isometry typically yields classification theorems. Approximate preservation yields stability theorems: in Hilbert $C^*$-modules, $(\delta,\varepsilon)$-orthogonality preservation forces quantitative closeness to a scaled inner-product preserving map, while in adaptive sensor coverage or positivity-preserving interpolation the preserved object is a constrained reduced model or constrained interpolant rather than an unrestricted linear operator (Moslehian et al., 2016, Shaffer et al., 12 Sep 2025, Ouermi et al., 2023).

2. Hilbert-space geometry and rigidity

In complex inner-product spaces, additive orthogonality preservation already imposes a strong metric form. If $H$ and $K$ are complex inner-product spaces with $\dim(H)\ge 2$, and $A:H\to K$ is additive and preserves orthogonality, then $A$ is zero or $A=\gamma T$ with $\gamma>0$ and $C^*$0 a real-linear isometry. The remaining ambiguity is precisely compatibility with multiplication by $C^*$1. The main classification states that $C^*$2 is complex-linear or conjugate-linear if and only if, for every $C^*$3, $C^*$4; equivalently, it is enough that this hold at one nonzero point, or that $C^*$5 for one nonzero $C^*$6. The same work shows that dense range, or finite-dimensionality with $C^*$7, rules out the genuinely mixed real-linear regime and forces $C^*$8 with $C^*$9 and $(\delta,\varepsilon)$0 a complex-linear or conjugate-linear isometry (Li et al., 20 Mar 2025).

In real inner-product spaces, the rigidity can be stated directly as similarity. A nonzero linear map $(\delta,\varepsilon)$1 is a similarity if and only if it preserves orthogonality, if and only if equal norms imply equal image norms, and if and only if order of norms is preserved. In that case $(\delta,\varepsilon)$2 and, by polarization, $(\delta,\varepsilon)$3. The same paper proves a fixed-angle variant: if $(\delta,\varepsilon)$4 is injective, nonzero, and preserves one angle $(\delta,\varepsilon)$5, together with the condition that equal-norm $(\delta,\varepsilon)$6-pairs map to equal-norm $(\delta,\varepsilon)$7-pairs, then $(\delta,\varepsilon)$8 is again a similarity (Moslehian et al., 2015).

A real-Hilbert-space analogue of Wigner-type rigidity appears in the classification of maps preserving the area of parallelograms. For $(\delta,\varepsilon)$9, a map $H$0 satisfying $H$1 has a dimension-dependent form. On $H$2, $H$3 with $H$4. In dimensions $H$5, $H$6 with $H$7 orthogonal. In infinite-dimensional real Hilbert spaces, bijective preservers are $H$8 with $H$9 a linear surjective isometry. The sign function is unavoidable because area is unchanged by $K$0 (Gehér, 2014).

3. Hilbert $K$1-modules, operator algebras, and cone geometry

For Hilbert $K$2-modules, orthogonality preservation admits both exact and approximate formulations. If $K$3 means $K$4, then a map $K$5 is $K$6-orthogonality preserving when $K$7. Under the standing assumption $K$8, a nonzero $K$9-linear $\dim(H)\ge 2$0-orthogonality preserving map satisfies

$\dim(H)\ge 2$1

and

$\dim(H)\ge 2$2

In particular, when $\dim(H)\ge 2$3, approximate preservation collapses to exact scaled inner-product preservation. A complementary exact characterization shows that, under the same algebraic inclusion, a nonzero $\dim(H)\ge 2$4-linear map is orthogonality preserving if and only if $\dim(H)\ge 2$5 for all $\dim(H)\ge 2$6 (Moslehian et al., 2016, Moslehian et al., 2015).

In $\dim(H)\ge 2$7-algebras, preservation of covariance sets and covariance cosets is tied to Moore–Penrose inversion. For $\dim(H)\ge 2$8, the covariance set is

$\dim(H)\ge 2$9

and an analogous definition gives the covariance coset. If $A:H\to K$0 has real rank zero, $A:H\to K$1 is prime, and $A:H\to K$2 is surjective, unital, linear, and satisfies $A:H\to K$3 for all $A:H\to K$4, then $A:H\to K$5 is either a $A:H\to K$6-homomorphism or a $A:H\to K$7-anti-homomorphism, and it preserves covariance sets and covariance cosets exactly: $A:H\to K$8 and $A:H\to K$9 (Alizadeh, 2018).

A cone-based analogue appears for the Lorentz spectrum on $A$0. If $A$1 denotes the set of $A$2 arising from the eigenvalue complementarity problem relative to the Lorentz cone, then every linear map $A$3 preserving $A$4 for all $A$5 has the form

$A$6

where $A$7 is an orthogonal $A$8 matrix. Thus Lorentz-spectrum preservation fixes the distinguished third coordinate and acts by a Euclidean isometry on the first two coordinates (Bueno et al., 2022).

4. Quantum states, channels, and multipartite operator spaces

In quantum mechanics, stochastic isometries provide one of the clearest formulations of a physics-preserving linear map. A stochastic map $A$9 on self-adjoint trace-class operators is positive and trace preserving; it is a stochastic isometry when it also preserves the trace norm. For positive operators, orthogonality is equivalent to $A=\gamma T$0, and a fundamental result states that a stochastic map is isometric if and only if it preserves orthogonality of states. Every stochastic isometry has the form

$A=\gamma T$1

where the $A=\gamma T$2 are unitary or antiunitary embeddings into mutually orthogonal subspaces and the $A=\gamma T$3 sum to $A=\gamma T$4. In the surjective case one recovers the Wigner–Kadison form $A=\gamma T$5, while complete positivity removes the antiunitary option and leaves only unitary pieces (Busch, 2013).

Preservation of pure states and separable pure states yields a multipartite preserver theory. For single systems, a linear pure-state preserver on $A=\gamma T$6 is either $A=\gamma T$7, where $A=\gamma T$8 is a fixed pure state and $A=\gamma T$9 vanishes on finite-rank operators, or $\gamma>0$0, where $\gamma>0$1 is a linear or conjugate-linear isometry; boundedness eliminates the singular $\gamma>0$2-term. For bipartite systems, preservation of separable pure states leads to nine possible forms, with the most physically standard being local isometric or conjugate-isometric product conjugations and swap-plus-local implementations. In the affine two-way classification, only the standard local forms remain, expressed through unitary or conjugate unitary maps together with identity, transpose, or partial transpose (Hou et al., 2012).

Tensor-product preserver problems on matrices give parallel results for local subsystem structure. If $\gamma>0$3 preserves the Ky Fan $\gamma>0$4-norm of every simple tensor, or the Schatten $\gamma>0$5-norm for $\gamma>0$6 with $\gamma>0$7, then

$\gamma>0$8

where $\gamma>0$9 are unitary and each $C^*$00 is either the identity or transpose map; the same pattern extends to $C^*$01 (Fosner et al., 2012). For Hermitian product tensors, preservation of spectrum or spectral radius yields the analogous unitary-similarity forms

$C^*$02

and, for spectral radius, an additional $C^*$03 (Fosner et al., 2012).

A broader channel-comparison framework replaces complete positivity by Hermitian-preserving trace-preserving linear maps. For finite-dimensional channels $C^*$04 and $C^*$05, the asymptotic statistical preorder $C^*$06 holds if and only if there exists a Hermitian-preserving trace-preserving map $C^*$07 such that $C^*$08; equivalently,

$C^*$09

The point of the construction is operational: $C^*$10 need not be positive, but it preserves Hermiticity and trace and exactly characterizes when the output of one channel can be reconstructed from the output statistics of another (Mitra et al., 8 Dec 2025).

5. Hamiltonian and numerical structure preservation

In accelerator physics, the relevant invariant is the symplectic form of Hamiltonian mechanics. If a transfer map sends entrance conditions $C^*$11 to exit conditions $C^*$12, then its Jacobian matrix $C^*$13 must satisfy

$C^*$14

The linear part $C^*$15 of a one-turn map is therefore a symplectic matrix, and higher-order corrections are constrained as well. A central construction is the Dragt–Finn type Lie factorization

$C^*$16

which preserves exact symplecticity term-by-term. The same chapter emphasizes a key limitation: truncating a Taylor expansion generally destroys exact symplecticity, so symplectic jets must be completed by Lie, generating-function, or Cremona-map constructions if long-time tracking is to remain physically faithful (Abell et al., 2022).

Some works broaden the term “mapping” beyond genuine linear operators. The “invertible linearization map” for the quartic oscillator is explicitly nonlinear: it combines a nonlinear algebraic change of coordinate with a nonlinear reparameterization of time, yet preserves potential energy, matches velocities and momenta, and gives a one-to-one correspondence between quartic-oscillator and harmonic-oscillator world lines. The author explicitly states that the map is not linear as a map of vector spaces; it is a linearizing transformation, not a linear operator (Anderson, 2012).

A similar qualification applies to positivity-preserving numerical transfer. HiPPIS constructs high-order data-bounded interpolation and positivity-preserving interpolation on structured meshes. For a fixed stencil, Newton interpolation is linear in the nodal values, but the overall map is nonlinear because the stencil is selected adaptively from data-dependent admissibility inequalities. The physically preserved object is intervalwise boundedness or positivity of quantities such as mass, density, and concentration, not global linearity of the transfer rule (Ouermi et al., 2023). Likewise, the CNWF digital twin for adaptive source localization uses linear FEEC components—basis pullbacks, mass matrices, and coboundary operators—inside an overall nonlinear sensor-to-source map constrained by a reduced conservation law (Shaffer et al., 12 Sep 2025).

6. Scope, limitations, and recurring distinctions

Several distinctions recur throughout the subject. First, preservation on a restricted test family is weaker than global preservation on the ambient space. Product-tensor norm or spectrum preservers are classified from their action on simple tensors, yet the cited papers explicitly note that such maps need not preserve the same invariant on arbitrary matrices (Fosner et al., 2012, Fosner et al., 2012). In accelerator physics, a truncated jet may agree with a symplectic map through a given order but still fail to be exactly symplectic under iteration (Abell et al., 2022). In interpolation, positivity at selected points is weaker than positivity of the interpolant over the whole interval, which is exactly the gap HiPPIS is designed to close (Ouermi et al., 2023).

Second, the word “physical” is not uniform across domains. In Hilbert-space geometry it usually means preserving orthogonality, angle, or transition structure. In operator algebra it can mean preserving covariance under generalized inversion. In quantum-state theory it means preserving positivity, trace, and trace-norm distinguishability, or preserving the extremal pure-state and separable-pure-state sets. In Hamiltonian mechanics it means exact preservation of the canonical symplectic structure. In numerical transfer it may mean preventing negative or otherwise inadmissible field values. The cited literature therefore does not reduce “physics-preserving linear mapping” to one canonical axiom system (Li et al., 20 Mar 2025, Busch, 2013, Abell et al., 2022, Ouermi et al., 2023).

Third, the boundary between linear, real-linear, conjugate-linear, antiunitary, and merely Hermitian-preserving behavior is often the decisive structural issue. Additive orthogonality preservers on complex Hilbert spaces are automatically real-linear isometries up to scale, but not automatically complex-linear or conjugate-linear (Li et al., 20 Mar 2025). Stochastic isometries allow antiunitary pieces unless complete positivity is imposed (Busch, 2013). Hermitian-preserving trace-preserving post-processings characterize a meaningful channel preorder even though they may fail positivity (Mitra et al., 8 Dec 2025). These distinctions explain why preserver theorems are typically sharp only after their hypotheses are stated with exceptional care.

Taken together, these results exhibit a common rigidity phenomenon. When the preserved relation is strong enough—orthogonality, exact distinguishability, product-operator spectral data, Lorentz-cone spectrum, covariance under Moore–Penrose inversion, or canonical symplecticity—the space of admissible transformations narrows dramatically. What remains are not arbitrary linear distortions, but highly structured maps: similarities, real-linear or complex-linear isometries, unitary or antiunitary embeddings, local identity/transpose actions on tensor factors, orthogonal block conjugations, or symplectic transfer maps. The broad subject of physics-preserving linear mapping is therefore less a single theory than a constellation of classification results unified by the same principle: preservation of a physically meaningful invariant is often far more restrictive than linearity itself.