Equivalence of Transformations: Theory & Applications

• Equivalence of transformations is the process of identifying distinct maps or actions as equivalent when they preserve specific invariants and structural properties.
• It applies to various fields such as differential equations, gauge theory, and tensor decompositions, providing a unified framework for analysis.
• A systematic approach using invariant coframes, groupoids, and canonical forms enables practical classification and consistent treatment of equivalence across disciplines.

Equivalence of transformations denotes a family of mathematical and physical notions in which distinct transformations, representations, or induced systems are regarded as the same when they preserve a specified structure, invariant, or observable content. In the literature gathered here, the preserved object ranges from Christoffel determinants and Green’s functions to induced maps on nearly clopen sets, classes of differential equations with arbitrary elements, pathwise program semantics, and polyadic decompositions of matrix multiplication tensors. The corresponding equivalence mechanisms include orthogonal similarity, conjugacy, Lie-group actions on arbitrary elements, deformation of integration domains, and Cartan-type invariant coframes (Bos et al., 2019, Rudolph et al., 2012, Cheviakov, 2017, Meerleer et al., 2019, Mueller-Hill et al., 2010, Kumpera, 2014).

1. Core meanings and recurring structures

Across the cited works, “equivalence” is not a single definition but a template: one fixes a class of admissible transformations and a notion of preserved structure, then asks when two transformed objects belong to the same orbit.

Setting	Transformations	Equivalence criterion
Christoffel equations	Orthogonal transformations of $p$ or $c$	$\check\Gamma=A^T\hat\Gamma A$ (Bos et al., 2019)
Nearly continuous dynamics	Conjugacy of induced maps on nearly clopen sets	Nearly continuous even Kakutani equivalence (Rudolph et al., 2012)
PDE classes with arbitrary elements	ETs and GETs	Every class member maps to another member (Cheviakov, 2017)
Array-intensive programs	Loop, expression, algebraic transforms	Same operator sequence and output-input mapping (0710.4689)
Matrix multiplication tensors	Permutations, scalings, trace transforms	Same tensor decomposition orbit (Berger et al., 2019)
Gauge-theory correlators	LKFTs and Nielsen identities	Same generalized Slavnov–Taylor identity (Meerleer et al., 2019)

A second recurring structure is the distinction between the equivalence group and the full set of admissible transformations. In the language of integrability and class analysis, admissible transformations form an equivalence groupoid, while normalized classes are those for which the entire groupoid is generated by the equivalence group (Vaneeva et al., 2013). This suggests a common organizational principle: equivalence problems are most tractable when admissible transformations admit a finite or canonical description in terms of invariants, normal forms, or orbit representatives.

2. Differential equations, arbitrary elements, and structure-preserving maps

For classes of differential equations with arbitrary elements, equivalence transformations are class-preserving point transformations acting simultaneously on independent variables, dependent variables, and arbitrary functions or parameters. A standard model is

$E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$

with global transformation law

$$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$

and infinitesimal invariance enforced by

$X^{(k)}E^\sigma=0$

on the solution manifold (Cheviakov, 2017). Generalized equivalence transformations extend this by allowing the variable transformation itself to depend on arbitrary elements. The cited paper gives a first computational example where the transformed dependent variable involves an arbitrary constitutive function: $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$ with

$$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$

which is strictly beyond ordinary point ETs (Cheviakov, 2017).

The same paper treats parameter reduction as a constructive use of equivalence. For anti-plane shear waves in incompressible hyperelastic fiber-reinforced media, the transformation

$$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$

maps any fiber angle $\gamma$ to the canonical direction $c$0 by choosing $c$1, thereby absorbing the fiber orientation into transformed coefficients $c$2 and then permitting nondimensional reduction to a parameter-free canonical equation (Cheviakov, 2017).

A complementary structural result is the “dummy variables” method for equivalence subgroups. If a simpler family has arbitrary elements depending on fewer variables, its equivalence group embeds into that of a more general family with larger dependence. Formally, if $c$3 has arbitrary elements $c$4 and $c$5 has the same structural equation with $c$6, then $c$7 under the hypotheses of Theorem 1 and Theorem 2 (Ndogmo, 2011). This provides a systematic reduction principle: compute the group for a simpler class, lift it, then impose dependence constraints.

For nonlinear wave-type equations of the form

$c$8

the admissible point transformations have the restricted form

$c$9

with $\check\Gamma=A^T\hat\Gamma A$0 projectable. A central obstruction is explicit: if $\check\Gamma=A^T\hat\Gamma A$1, then there is no admissible point transformation mapping nonlinear members to linear ones; at least one of $\check\Gamma=A^T\hat\Gamma A$2 must depend on $\check\Gamma=A^T\hat\Gamma A$3 (Özer, 2018). The examples

$\check\Gamma=A^T\hat\Gamma A$4

and its two-variable analogues show how nonlinear equations can be linearized and exact solutions imported from the linear wave equation (Özer, 2018).

Equivalence can also be used to remove explicit sources and nonautonomous terms. A class of one conservation law and four balance laws in $\check\Gamma=A^T\hat\Gamma A$5-dimensions, depending on sixteen arbitrary functions $\check\Gamma=A^T\hat\Gamma A$6, is studied in an augmented space $\check\Gamma=A^T\hat\Gamma A$7. By projecting admitted symmetries and integrating the resulting generators, one obtains finite transformations mapping the original system to an equivalent autonomous system of conservation laws (Gorgone et al., 2021). In the physical application to the 3D Euler equations in a rotating, gravitational frame, the transformation eliminates Coriolis, centrifugal, and gravity terms from the target representation, while preserving the differential structure (Gorgone et al., 2021).

3. Tensorial, geometric, and variational equivalence

In anisotropic wave propagation, the equivalence problem appears as an orthogonal-similarity statement for Christoffel matrices. For a fourth-rank tensor $\check\Gamma=A^T\hat\Gamma A$8, vector $\check\Gamma=A^T\hat\Gamma A$9, and $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$0, define

$E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$1

Then the associated Christoffel-type matrices

$E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$2

satisfy

$E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$3

Hence

$E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$4

so rotating $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$5 with fixed $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$6 and rotating $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$7 with fixed $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$8 are equivalent determinantal forms. The paper explicitly emphasizes the non-equivalence of these rotated forms with the untransformed determinant in general (Bos et al., 2019).

The local and global equivalence of geometric structures is formulated on prolongation spaces $E^\sigma\bigl(x,u,u_{(k)};a,f\bigr)=0,$9, where a structure is a section $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$0. Finite equivalence jets are encoded by

$$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$1

and infinitesimal equivalence by

$$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$2

The Cartan–Ehresmann–Spencer framework then studies regularity, subgroupoid structure, and differential invariants via prolongation, symbol calculus, and Spencer $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$3-cohomology (Kumpera, 2014). In this setting, equivalence is governed not by a single scalar criterion but by a hierarchy of jet-level orbit conditions and finite generating systems of invariants.

A deeper triadic equivalence occurs for variational problems of order $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$4: the equivalence problem for higher-order Lagrangians modulo contact transformations, nonzero scalar multiplication, and divergence; the equivalence problem for associated rank-2 distributions defined by underdetermined ODEs $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$5; and the equivalence problem for variational ODEs of order $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$6 are “essentially the same” (Doubrov et al., 2010). Their common geometry is the linearization to self-dual curves in odd-dimensional projective spaces, with maximal symmetry characterized by rational normal curves. For the maximally symmetric model,

$$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$7

all generalized Wilczynski invariants vanish (Doubrov et al., 2010).

Cartan-type operator equivalence appears in scalar differential operators on the line. For third-order operators, the equivalence problem under fiber-preserving transformations $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$8, $$x' = X(x,u;\varepsilon),\qquad u' = U(x,u;\varepsilon),\qquad a' = A(a;\varepsilon),\qquad f' = F(x,u,f;\varepsilon),$$9 is solved in two versions: direct equivalence and gauge equivalence. Canonical invariant coframes lead to structure equations whose invariants $X^{(k)}E^\sigma=0$0 or $X^{(k)}E^\sigma=0$1 completely characterize equivalence classes (Nadjafikhah et al., 2011). The fourth-order analogue treats

$X^{(k)}E^\sigma=0$2

under

$X^{(k)}E^\sigma=0$3

again distinguishing direct and gauge equivalence, with complete classification in terms of invariant coframes and fundamental invariants (Bakhshandeh-Chamazkoti, 2014).

For generalized Abel ODEs under the linear pseudogroup

$X^{(k)}E^\sigma=0$4

local equivalence is likewise reduced to differential invariants and invariant derivations. In the classical cubic case,

$X^{(k)}E^\sigma=0$5

the algebra of differential invariants is generated by one absolute invariant $X^{(k)}E^\sigma=0$6 and one invariant derivation $X^{(k)}E^\sigma=0$7, and two regular equations are locally equivalent iff the corresponding invariant curves coincide (Jr, 2014).

4. Gauge dependence, domain deformation, and physical equivalence

In non-Abelian gauge theory, Landau–Khalatnikov–Fradkin transformations and Nielsen identities give two descriptions of gauge-parameter dependence within linear covariant gauges. The paper constructs the gauge-invariant composite field

$X^{(k)}E^\sigma=0$8

and uses an extended BRST symmetry

$X^{(k)}E^\sigma=0$9

to derive a generalized Slavnov–Taylor identity from which both LKFTs and Nielsen identities follow (Meerleer et al., 2019). The resulting equivalence is exact: finite gauge transport by LKFTs and differential gauge transport by Nielsen identities are two manifestations of the same BRST-generated flow. The paper also gives an explicit one-loop LKFT for the gluon propagator and reproduces its $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$0-dependence, including the undressed longitudinal sector (Meerleer et al., 2019).

In the theory of disordered electron systems, equivalence of transformations is realized by deformation of integration domains for hyperbolic Hubbard–Stratonovich transformations. The Pruisken–Schäfer domain, the Schäfer–Wegner domain, and the Euclidean Gaussian domain are connected by explicit homotopies, and the central Gaussian identity is shown to persist under these deformations for general non-compact symmetry groups when $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$1 (Mueller-Hill et al., 2010). The same analysis explains the sign-weighted Jacobian for orthogonal symmetry; in the $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$2 case,

$$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$3

while no alternating sign appears for $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$4, where the Vandermonde is squared (Mueller-Hill et al., 2010).

In scalar–tensor EFT, Weyl-related frames are physically equivalent only when contact terms induced by graviton exchange are retained. Starting from a non-minimal coupling $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$5, the direct diagrammatic analysis produces local operators

$$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$6

which coincide with the operators induced by a Weyl transformation to the minimal Einstein–Hilbert frame (Hill et al., 2020). The result is framed as an explicit realization of the equivalence theorem: Jordan and Einstein descriptions are classically and quantum mechanically equivalent in the EFT regime, provided the hidden contact terms and Jacobians are included (Hill et al., 2020).

A nonequilibrium thermodynamic analogue appears in diffusive systems within macroscopic fluctuation theory. Quasistatic protocols $$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$7 are called equivalent when they connect the same stationary endpoints and yield the same free-energy difference, via the exact identity

$$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$8

independent of the chosen quasistatic path (Gabrielli et al., 17 Sep 2025). Equivalent quasistatic transformations are nevertheless distinguishable at finite speed by the renormalized work,

$$x_* = F_1(t)\,x + F_0(t),\qquad t_* = G(t),\qquad u_* = U_1(t)\,u + U_0(x,t),$$9

so optimality is defined by minimizing the excess functional $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$0 among equivalent protocols (Gabrielli et al., 17 Sep 2025).

5. Dynamical, algorithmic, and tensor-computational equivalence

In ergodic theory, Kakutani equivalence compares induced transformations rather than full systems. For nearly continuous systems, nearly continuous even Kakutani equivalence requires nearly clopen sets $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$1, $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$2 with $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$3 and a near homeomorphism $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$4 satisfying

$$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$5

A major result is that such a conjugacy extends to a nearly continuous orbit equivalence of the ambient systems, so the definition based only on induced maps coincides with the earlier del Junco–Rudolph–Weiss formulation involving orbit equivalence on full-measure sets (Rudolph et al., 2012). The same paper shows that strongly rank one transformations, including Chacon’s map, lie in the same nearly continuous even Kakutani equivalence class as irrational rotations (Rudolph et al., 2012).

For array-intensive source programs, functional equivalence is formalized as

$$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$6

or elementwise equality on output arrays (0710.4689). The method applies to a restricted but practically relevant program class: dynamic single-assignment form, static control flow, affine or piecewise affine index expressions, and no pointers. Each program is translated into an Array Data Dependence Graph, and equivalence is certified when every corresponding path has the same operator sequence and the same output-input mapping. Associativity and commutativity are handled by flattening and matching. The result is a sufficient, sound condition rather than a complete decision procedure; the implementation reports diagnostics when mappings disagree, and reported runtimes are consistently under 100 seconds for realistic kernels (0710.4689).

For matrix multiplication tensors

$$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$7

equivalence of polyadic decompositions is defined by three families of invariance transformations: permutations of rank-1 terms, per-term scalings $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$8 with $$U_0(x,t)=\frac{F_1'(t)\,x+F_0'(t)}{F_1(t)\,C(x,t)}\,U_1(t),$$9, and trace transformations

$$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$0

The paper develops an algorithm based on simultaneous similarity of the signatures $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$1, clustering numbers, and linear recovery of the $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$2-parameters (Berger et al., 2019). It confirms de Groote’s theorem that all rank-7 decompositions of $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$3 are equivalent to Strassen’s algorithm, but finds that decompositions for larger tensors such as $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$4 by $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$5 and $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$6 by $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$7 are very likely to be essentially different. It also gives a necessary criterion for equivalence to decompositions with integer or power-of-two entries, relevant for stability and efficiency (Berger et al., 2019).

6. Invariants, obstructions, and the limits of equivalence

A persistent theme is that equivalence is rarely identity of formulas; it is identity modulo an admissible transformation class, and the admissible class matters. In normalized PDE classes, admissible transformations are generated by the equivalence group, but in non-normalized classes there exist form-preserving transformations not captured by the ordinary equivalence group (Cheviakov, 2017). The same distinction appears in the equivalence groupoid approach to integrability, where classification modulo equivalence removes inessential arbitrary elements and identifies canonical representatives (Vaneeva et al., 2013).

Several papers explicitly warn against overextending equivalence statements. For Christoffel equations, the equivalence proven is only between the two rotated determinant forms; it does not imply equivalence with

$$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$8

in general, and orthogonality is essential because a general $$G^* = G - S\,x,\qquad \tan\gamma^* = \tan\gamma + S$$9 change of basis does not preserve the relevant Euclidean structure (Bos et al., 2019). In nearly continuous Kakutani theory, “evenness” $\gamma$0 is crucial for extending conjugacy of induced maps to orbit equivalence, while the non-even case requires additional hypotheses such as nearly unique ergodicity (Rudolph et al., 2012). For program equivalence, the ADDG criterion is sufficient but not necessary, floating-point non-associativity is excluded, and pointer aliasing is disallowed (0710.4689).

The same pattern recurs in physics. Weyl-frame equivalence requires the contact terms generated by graviton exchange and the appropriate Jacobian accounting; without them, frame dependence is only apparent (Hill et al., 2020). Hyperbolic Hubbard–Stratonovich equivalence requires the correct domain deformation and the positivity hypothesis $\gamma$1 (Mueller-Hill et al., 2010). This suggests that equivalence of transformations is best understood not as formal transformability alone, but as transformability together with the preservation of the full invariant data: spectra, orbit structure, exact forms, Jacobian signs, or observable amplitudes.

Taken together, these works show that equivalence of transformations is a technical language for identifying when distinct descriptions belong to the same mathematical or physical orbit. Its operative content lies in the invariants that survive transformation—determinants, spectra, induced-return structures, classifying coframes, gauge-independent correlators, exact functionals, or input-output semantics—and in the precise specification of what is allowed to change and what must remain fixed.