Canonical Forms: Theory & Applications

• Canonical forms are unique representatives chosen from equivalence classes generated by group actions, encoding the essential structure of mathematical objects.
• They are constructed using diverse methodologies such as matrix decompositions, partition-refinement, and residue calculus to ensure computational tractability and uniqueness.
• Applications span pure and applied mathematics including coding theory, quantum information, and tensor networks, illustrating both theoretical depth and practical utility.

Canonical forms are distinguished, canonical representatives in equivalence classes induced by a group action, algebraic relation, or combinatorial structure. The canonical form framework serves as a cornerstone across linear algebra, algebraic geometry, combinatorics, invariant theory, coding theory, quantum information, and applied mathematics. The utility of canonical forms lies in their capacity to uniquely (or nearly so) encode the essential structure of complex mathematical objects up to an equivalence, enabling classification, computation, and invariant extraction.

1. Classification and Invariant Theory: Purpose and Examples

Canonical forms provide a rigorous method for classifying mathematical objects up to an equivalence—most often equivalence under some group action, such as similarity, congruence, or action by automorphisms. For a set $X$ equipped with an action by a group $G$, a canonical form is typically a map $\mathrm{Can}_G: X \to X$ such that for all $g \in G$, $\mathrm{Can}_G(g \cdot x) = \mathrm{Can}_G(x)$, and for each equivalence class, a unique representative is distinguished.

Classical canonical forms include:

• Jordan canonical form: for classifying complex matrices up to similarity.
• Schur form: for unitary similarity.
• Smith and Frobenius forms: for modules over principal ideal domains.
• Belitskii, Kronecker, or Weierstrass forms: for matrix pencils and system theory (Chen et al., 2017).
• Canonical forms for *-congruence and real congruence: for quadratic forms and bilinear forms (Terán et al., 8 Oct 2025).

Applications extend to:

• Coding theory: canonical forms distinguish codes up to semilinear equivalence (Feulner, 2013).
• Quantum information: SLOCC canonical forms for multi-qubit quantum states (Sudha et al., 2020).
• Tensor networks: gauge/canonical forms for optimizing numerical stability (Zhang et al., 2020).

2. Unique Canonical Forms: Existence, Computation, and Uniqueness

The existence and construction of canonical forms depend on the algebraic or combinatorial structure under consideration, often supported by a finiteness or reducibility property of the acting group. Uniqueness (up to prescribed ambiguities) is ensured either by algebraic invariants (e.g., eigenvalues, elementary divisors) or by structure theorems yielding normal forms.

• For nonderogatory matrices (characteristic polynomial equals minimal polynomial), a unique canonical form up to unitary similarity is obtained as a block-upper-triangular matrix governed by an acyclic undirected graph. Two such matrices are unitarily similar if and only if their associated canonical forms coincide (Futorny et al., 2011).
• Polynomial canonical forms (e.g., mixed sum-of-powers decompositions) are explicitly constructed by dimension counting and Jacobian verification, yielding uniqueness on dense open subsets (Reznick, 2012).
• In systems theory, Belitskii’s reduction algorithm yields canonical forms for dynamical systems under strict equivalence, with indecomposable classes built constructively by successive block reductions (Chen et al., 2017).
• In semigroup theory, canonical form algorithms based on a set of elementary identities enable a unique representative for each κ-term over free semigroups (Costa, 2013).

3. Algorithmic Approaches and Complexity

The computation of canonical forms often relies on algorithmic frameworks tailored to the group action or combinatorial context:

• Partition-refinement and individualization algorithms: used to canonize objects under permutation or semilinear actions, e.g., Feulner’s extension of the nauty framework to coding theory and projective spaces (Feulner, 2013).
• Graph-based normalization: phase-normalization algorithms for unitary similarity of nonderogatory matrices are tracked via acyclic graphs, which record dependencies between phase choices (Futorny et al., 2011).
• Algebraic and combinatorial pruning: in canonical forms of neural ideals, a polynomial-time algorithm using local combinatorial checks replaces expensive primary decompositions (Geller et al., 2022).
• Quantum circuits: canonical form construction for single-qutrit Clifford+T operators operates in linear time in the number of T gates, using rewriting rules and adjoint representations (Glaudell et al., 2018).
• Tensor networks: efficient canonicalization (e.g., QR sweeps) enables optimal conditioning for matrix product states but is computationally intensive for higher-dimensional PEPS (Zhang et al., 2020).
• Systems and polynomials: explicit dimension-counting and recursive procedures are used for existence and uniqueness proofs (Reznick, 2012, Chen et al., 2017).

A recurring theme is the trade-off between the expressivity of the group action (e.g., semilinear, local, or global transformations) and the computational complexity of finding canonical forms—often GI-hard in the presence of symmetries.

4. Canonical Forms in Positive Geometries and Canonical Differential Forms

In the geometric setting, canonical forms arise as rational differential forms with characteristic factorization and residue properties. Positive geometries, as defined in projective, Grassmann, or flag varieties, carry canonical forms characterized by:

• Simple pole singularities only on boundaries.
• Residue along boundary components equals the canonical form on that component (Arkani-Hamed et al., 2017).

For convex polytopes, the canonical form is

$\Omega_P = \frac{A(x)}{\prod_{j=1}^d L_j(x)} dx,$

where $L_j(x)$ are facet equations and $A(x)$ is the adjoint of the dual cone, cancelling spurious poles (Gaetz, 9 Apr 2025). Canonical forms admit dual volume representations as Laplace transforms of positivity-preserving measures supported on the dual cone, with complete monotonicity controlling positivity and uniqueness (Mazzucchelli et al., 2 Sep 2025). In the context of scattering amplitudes and amplituhedra, canonical forms encode important physical and combinatorial structures (Arkani-Hamed et al., 2019, Arkani-Hamed et al., 2017).

The unifying theme is that canonical forms in positive geometries serve as volume forms or generating functions that are uniquely determined by their prescribed singularities and recursion relations, capturing the essential geometry and enabling residue calculus, triangulation, and pushforward techniques (Arkani-Hamed et al., 2017, Gaetz, 9 Apr 2025, Mazzucchelli et al., 2 Sep 2025). Stringy canonical forms interpolate between polytope canonical forms and string integrals, preserving simple pole structure and encoding field theory limits (Arkani-Hamed et al., 2019).

5. Canonical Forms for Group Actions: Coding Theory, Neural Ideals, and Semigroups

In discrete mathematics and combinatorics, canonical forms under group actions underlie efficient classification and invariant computations:

• Coding theory: Canonical forms for codes and subspace configurations under semilinear group actions are crucial for isomorphism testing, orbit enumeration, and automorphism group computation. This is done via partition-refinement algorithms and projective semilinear actions (Feulner, 2013).
• Neural ideals: Canonical forms for pseudomonomial ideals classify the minimal relations among neural codewords and support efficient extraction of convexity obstructions, employing polarization and local minimal generator checks (Geller et al., 2022).
• Free κ-semigroups: Canonical forms are obtained by systematic application of pseudoidentities, generating a decidable word problem and unique normal forms for ω-terms over the variety of finite semigroups (Costa, 2013).

These structures rely on the identification and resolution of minimal generating sets, automorphism groups, and the construction of canonical representatives under permutation, semilinear, or algebraic operations.

6. Canonical Forms in Quantum Information and Tensor Networks

Canonical forms provide classification and computational tractability in high-dimensional and quantum contexts:

• Two-qubit SLOCC equivalence: Each two-qubit density matrix can be brought to one of two canonical forms under stochastic local operations: the Bell-diagonal and a non-diagonal rank-deficient form, utilizing Lorentz-congruence and Minkowski metrics for invariant characterization (Sudha et al., 2020).
• Single-qutrit Clifford+T: A unique, T-optimal canonical form exists for each single-qutrit operator, and the normal form can be computed efficiently in terms of "syllable" structure and T-count (Glaudell et al., 2018).
• Tensor network canonicalization: In matrix product and projected entangled pair states, (block)-isometric canonical forms minimize numerical error and condition number, crucial for stability, compressibility, and local optimization. For MPS the (mixed) canonical form is achieved via QR or SVD sweeps, whereas for PEPS only approximate column- or row-wise canonicalization is practical (Zhang et al., 2020).

In these contexts, canonical forms not only classify but optimize, enabling stable computation, optimal resource usage, and mapping to invariant subspaces.

7. Canonical Forms under Congruence and Broader Matrix Theories

Matrix congruence and *-congruence canonical forms play a central role in the classification of bilinear and sesquilinear forms:

• Real congruence canonical forms: Every real matrix is congruent to a direct sum of blocks of four (or, in tridiagonal form, three) explicit types—nilpotent Jordans, signature+rotation blocks, Hermitian-cosquare, and companion-cosquare blocks—fully determined by the structure of elementary divisors and cosquare invariants (Terán et al., 8 Oct 2025).
• Relation to Kronecker canonical form: The block structure is dictated by real Kronecker invariants of the matrix pencil $(A^T, A)$, up to certain sign ambiguities.
• Comparison to Lee–Weinberg form: The block types of diverse canonical forms are explicitly matched, clarifying redundancies and refining classification (Terán et al., 8 Oct 2025).

These results illustrate the continued significance of canonical forms in matrix theory, system theory, and the algebraic analysis of linear operators.

---

Canonical forms underpin a spectrum of algorithmic, algebraic, geometric, and physical applications. Their construction depends critically on the nature of the equivalence considered and the algebraic/geometric objects studied. The wide variety of approaches—from combinatorial partition-refinement to residue calculus and Laplace duality—emphasize the depth and utility of canonical forms as organizing principles in both pure and applied mathematics.