Representation-Protected Invariants

• Representation-protected invariants are invariant properties defined through a representation object that constrains state transformations rather than direct state access.
• They integrate techniques from program verification, topological band theory, and higher algebra to enforce and detect deformations via structured representations.
• Applications span automated software verification, topological material design, and categorical knot homology, offering robust invariance in complex systems.

"Representation-protected invariants" (Editor's term) denotes a family of invariant constructions in which the decisive information is carried by a representation object—such as a hidden program representation, a symbolic automaton of library interactions, a symmetry sewing matrix, a partial symmetry operator, a categorical representation, or a group representation—and in which preservation or detectability is mediated by that representation rather than by unrestricted access to underlying state. Across recent work, this viewpoint appears in modular program verification, topological band theory, interacting and crystalline topological phases, higher representation theory, modular representation theory, and combinatorial representation data (Zhou et al., 2024, Ahn et al., 2018, Shiozaki et al., 2016, Müller et al., 2022).

1. Cross-domain structure of the notion

A common pattern is that an invariant is not introduced as a bare property of a state space, but as a property of how that state is represented, transformed, or hidden. In software verification, the invariant concerns hidden state and is enforced through admissible access paths, automata, ownership, or abstract interpretation. In topological matter, the invariant is encoded in a symmetry representation matrix, a defect representation, or a partial symmetry expectation value. In algebra and categorification, the invariant is singled out by a representation category, a Grothendieck-group decategorification, or a fixed-vector subspace of a group action. This suggests a unifying schema in which the representation carries the invariant and simultaneously constrains the admissible deformations (Zhou et al., 2024, Ahn et al., 2018, Webster, 2013).

Domain	Representation object	Invariant form
Program verification	HAT/SFA, ghost ownership state, MRU object abstraction	Representation invariant over hidden state
Band topology and superconductivity	Sewing matrix $G(\mathbf k)$, symmetry sector decomposition	Homotopy class, $w_2$, $\mathbb Z_2$ crystalline invariant
Interacting topological phases	Partial symmetry operator, defect data, $G$-crossed BTC data	Quantized phase, defect charge/statistics, crystalline invariant
Categorification and modular representation theory	Tensor-product categories, Weil representation	Knot homology, invariant subspace
Partition data	Pair of partitions $(\lambda',\lambda'')$	Symbol invariant, fingerprint invariant

The strongest versions of the idea are those in which triviality is formulated as the ability to deform a representation to a canonical one. In $C_{2z}T$-symmetric band theory, trivial topology is equivalent to deforming the sewing matrix to a constant matrix $G_0$ independent of $\mathbf k$ (Ahn et al., 2018). In functional verification over hidden libraries, the safe executions are those whose traces stay inside an automaton $A_{\mathit{inv}}$ recognizing exactly the histories that preserve the representation invariant (Zhou et al., 2024). In the Weil-representation setting, invariant vectors are those fixed by $\rho_D(\mathrm{SL}_2(\mathbb Z))$, and all such invariants are generated from five fundamental ones by isotropic induction (Müller et al., 2022).

2. Hidden representations in software verification

In verification of functional programs over stateful libraries, the central problem is how to statically guarantee representation invariants for functional data structures whose implementations rely on stateful library calls and hidden internal state. The invariant is not merely a predicate over a single memory snapshot: it is often a property of the implementation’s observable interaction trace with the hidden stateful library, and it must hold after every relevant stateful computation. "A HAT Trick" introduces symbolic finite automata (SFA) as a history language for these interactions and integrates them into a refinement type system via Hoare Automata Types (HATs). The core judgments have the shape

$w_2$0

and invariant preservation is organized by defining an automaton $w_2$1 that recognizes exactly the histories that preserve the representation invariant. The bidirectional algorithm then generates HAT inclusion obligations and reduces them to SMT-friendly verification conditions (Zhou et al., 2024).

This automata-and-types viewpoint is complemented by semantic collaboration for sequential object-oriented programs. "Flexible Invariants Through Semantic Collaboration" treats representation-protected invariants as invariants that depend not only on an object’s own representation, but also on selected collaborating objects. The methodology introduces built-in ghost attributes closed, owns, owner, subjects, and observers, and proves global validity through

$w_2$2

together with ownership consistency

$w_2$3

Admissibility requires, among other conditions, that

$w_2$4

that subjects know their observers, and that updates satisfy the guard-preservation condition. The method is implemented in AutoProof and was evaluated on Observer, Iterator, Master clock, Doubly-linked list, Composite, and PIP (Polikarpova et al., 2013).

Automatic inference of such invariants has recently been addressed by abstract interpretation. "Automatic Inference of Relational Object Invariants" studies relational object invariants for heap-allocated objects in C-like programs and introduces the recent-use memory model (RUMM), in which each memory bank contains storage, a cache representing the most recently used (MRU) object, and bookkeeping flags. Its abstract domain, MRUD, is a reduced product of a numerical domain, an equality domain, a pointer/base-address equality domain, and a banked object domain. The design allows strong updates on the cached object while preserving the summary invariant for the rest of the bank, which is crucial because the invariant may be temporarily broken during field updates. The implementation in Crab reports that, with Zones as the numerical domain, MRUD is about 75x faster on average than the state-of-the-art summarization-based implementation, that average speedups are about 81x, 76x, and 57x depending on reduction strategy, and that MRUD proves all assertions on the precision benchmark suite (Su et al., 2024).

3. Symmetry representation as topological invariant

In $w_2$5-symmetric band systems, the representation itself becomes the topological carrier. The central object is the sewing matrix

$w_2$6

which transforms under gauge change as

$w_2$7

When the Bloch states are chosen smooth over the Brillouin zone, $w_2$8 is also smooth, and nontrivial band topology is equivalent to an obstruction to deforming $w_2$9 to a constant matrix $\mathbb Z_2$0 independent of $\mathbb Z_2$1. On a $\mathbb Z_2$2-invariant plane the symmetry constraint forces

$\mathbb Z_2$3

so the topology is that of a map into $\mathbb Z_2$4. The paper proves that the second Stiefel-Whitney number $\mathbb Z_2$5 is precisely the homotopy invariant of $\mathbb Z_2$6, relates $\mathbb Z_2$7 to the Wilson-loop winding, and identifies the 3D strong invariant

$\mathbb Z_2$8

with the quantized magnetoelectric polarizability through

$\mathbb Z_2$9

The resulting 3D strong Stiefel-Whitney insulator supports both 2D massless surface Dirac fermions and 1D chiral hinge states (Ahn et al., 2018).

A related but distinct representation-protected mechanism appears in time-reversal-invariant 3D topological superconductors with crystalline symmetry. A space-group symmetry $G$0 block-diagonalizes the BdG Hamiltonian into sectors $G$1, and a one-dimensional $G$2 invariant $G$3 protects surface Majorana Kramers pairs (MKPs). The classification depends on whether $G$4 or $G$5, and the magnetic response depends on whether the surface hosts one MKP or two MKPs. The paper identifies three response types (Yamazaki et al., 2019).

Type	Symmetry / sector data	Magnetic response
(A)	$G$6, $G$7, class D, one MKP	uniaxial anisotropy
(B)	$G$8, $G$9, class DIII, one MKP	Ising-like linear anisotropy
(C)	$(\lambda',\lambda'')$0, $(\lambda',\lambda'')$1, class DIII, two MKPs	biaxial / quadrupolar anisotropy

For Type (C), the surface gap takes the form

$(\lambda',\lambda'')$2

which yields the paper’s "biaxial / quadrupolar anisotropic magnetic response" (Yamazaki et al., 2019).

4. Interacting topological matter, defects, and partial symmetry probes

In interacting symmetry-protected topological phases, the abstract invariant is a cohomology class, but the physically useful invariant is a quantized response. For bosons in $(\lambda',\lambda'')$3-dimensional spacetime with on-site symmetry $(\lambda',\lambda'')$4, the classification is

$(\lambda',\lambda'')$5

with topological partition function

$(\lambda',\lambda'')$6

The paper defines SPT invariants as quantized physical responses to a background symmetry gauge configuration and emphasizes defect charge, defect statistics, boundary projective representations, induced lower-dimensional SPT states, and quantized topological response terms after gauging. A standard example is the $(\lambda',\lambda'')$7D bosonic $(\lambda',\lambda'')$8 SPT classified by $(\lambda',\lambda'')$9, for which $C_{2z}T$0 identical elementary monodromy defects carry total $C_{2z}T$1 charge $C_{2z}T$2 and the elementary monodromy defect has statistical angle

$C_{2z}T$3

A many-body generalization of the same logic uses partial point-group transformations. For a spatial subregion $C_{2z}T$4 closed under the symmetry action, the partial transformation $C_{2z}T$5 yields a ground-state overlap

$C_{2z}T$6

Here $C_{2z}T$7 is the universal topological $C_{2z}T$8 phase, $C_{2z}T$9 is a scale-independent contribution to the amplitude, and the phase is quantized because the overlap is interpreted as a TQFT partition function on a spacetime manifold determined by the partial symmetry operation. In the $G_0$0D reflection-symmetric Kitaev chain,

$G_0$1

which generates the $G_0$2 classification. In $G_0$3D inversion-symmetric class D superconductors, partial inversion on a 3-ball gives

$G_0$4

realizing the $G_0$5 root phase (Shiozaki et al., 2016).

Crystalline invariants of fractional Chern insulators extend this partial-symmetry strategy to symmetry-enriched topological order. For a large disk-like subregion $G_0$6 centered at a high-symmetry point $G_0$7, the partial rotation expectation value is

$G_0$8

Using conformal field theory and $G_0$9-crossed braided tensor categories, the paper shows that, for the topological orders considered, the Hall conductivity, filling fraction, and partial rotation invariants fully characterize the crystalline invariants of the system. For the square lattice,

$\mathbf k$0

linking rotation centers, anyon per unit cell, and defect responses (Kobayashi et al., 2024).

Non-symmorphic semimetals supply a complementary symmetry-response invariant. Adapting Oshikawa’s flux-insertion argument to a non-symmorphic space-group operation $\mathbf k$1, the paper derives a generalized topological Luttinger invariant

$\mathbf k$2

Unlike ordinary Luttinger’s theorem, this invariant can remain nonzero even when the Fermi sea volume vanishes. A nonzero Luttinger invariant then forces energy bands to touch, and opening a gap without symmetry breaking automatically triggers fractionalization (Parameswaran, 2015).

5. Higher representation theory, fixed vectors, and combinatorial representatives

In higher representation theory, knot homology is literally built from representation categories. Webster constructs a homology theory $\mathbf k$3 for links $\mathbf k$4 whose components are labeled by finite-dimensional representations of a simple complex Lie algebra $\mathbf k$5, with graded Euler characteristic equal to the corresponding quantum invariant. The tensor-product category satisfies

$\mathbf k$6

and the categorified braid group action is implemented by functors

$\mathbf k$7

The sequel constructs functors corresponding to braiding and the (co)evaluation maps between representations of quantum groups, defines $\mathbf k$8 for a labeled tangle $\mathbf k$9, and proves that $A_{\mathit{inv}}$0 depends only on the isotopy class of $A_{\mathit{inv}}$1. In this setting, the invariant is representation-protected because the labels are built into a categorical ribbon structure and diagrammatic moves are implemented by categorified representation-theoretic equivalences (Webster, 2013, Webster, 2010).

A fixed-vector version of the same theme appears in the Weil representation of $A_{\mathit{inv}}$2. For a discriminant form $A_{\mathit{inv}}$3, the invariant space is

$A_{\mathit{inv}}$4

with projection operator

$A_{\mathit{inv}}$5

If $A_{\mathit{inv}}$6 is isotropic, the isotropic lift

$A_{\mathit{inv}}$7

commutes with the Weil representation and sends invariants to invariants. The main theorem states that every invariant for a discriminant form of prime-power level is generated by isotropic lifts of one of five fundamental invariants $A_{\mathit{inv}}$8, and if $A_{\mathit{inv}}$9, then

$\rho_D(\mathrm{SL}_2(\mathbb Z))$0

The same classification yields explicit generators of Jacobi forms of singular weight (Müller et al., 2022).

Combinatorial representation data furnish another variant. For rigid semisimple surface operators in the classical theories $\rho_D(\mathrm{SL}_2(\mathbb Z))$1, $\rho_D(\mathrm{SL}_2(\mathbb Z))$2, and $\rho_D(\mathrm{SL}_2(\mathbb Z))$3, a pair of partitions $\rho_D(\mathrm{SL}_2(\mathbb Z))$4 carries two invariants: the symbol invariant and the fingerprint invariant. The symbol is computed from the partition data by forming $\rho_D(\mathrm{SL}_2(\mathbb Z))$5, separating odd and even subsequences, and defining

$\rho_D(\mathrm{SL}_2(\mathbb Z))$6

The fingerprint is defined through the intermediate partition $\rho_D(\mathrm{SL}_2(\mathbb Z))$7, where $\rho_D(\mathrm{SL}_2(\mathbb Z))$8, together with the sign function $\rho_D(\mathrm{SL}_2(\mathbb Z))$9. The paper constructs a "nice representative element" $w_2$00 with the same symbol invariant and shows that the fingerprint of the representative element can be obtained immediately. It also constructs representative elements with the same fingerprint invariant, thereby giving explicit maps between the two invariants (Shou et al., 2017).

6. Stability conditions, misconceptions, and limits

A recurring misconception is that a representation invariant must be a local predicate on an instantaneous state. Several of these works make the opposite point. In HATs, the invariant is a property of fine-grained temporal and data-dependent histories of interactions between functional clients and stateful libraries; in MRUD, relational object invariants are particularly challenging because they are often broken temporarily during field updates; and in semantic collaboration, an object’s consistency may depend on collaborators rather than only on a representation tree (Zhou et al., 2024, Su et al., 2024, Polikarpova et al., 2013).

Another misconception is that once a representation has been identified, the resulting invariant is automatically stable under arbitrary extensions. The band-topology literature states explicit stability conditions: the 3D strong invariant $w_2$01 is a well-defined stable bulk invariant only when all bulk Chern numbers vanish, and $w_2$02 can be fragile when $w_2$03 is nontrivial. The non-symmorphic Luttinger invariant is derived for spinless fermions or spinful fermions with conserved spin rotation and does not directly apply to generic spin-orbit coupled systems lacking spin-rotation symmetry (Ahn et al., 2018, Parameswaran, 2015).

A further point is that the correct invariant may require a partial rather than a global symmetry operation. In $w_2$04D reflection-symmetric superconductors, a full reflection on twisted ground states only captures a $w_2$05 subgroup via the Klein bottle, whereas the full $w_2$06 invariant requires partial reflection on an interval. In $w_2$07D bosonic $w_2$08 SPT phases, total defect charge distinguishes the $w_2$09 phases fully only for odd $w_2$10; for even $w_2$11, one needs the defect statistics invariant to fully separate them (Shiozaki et al., 2016, Wen, 2013).

Finally, representation-protected constructions are not uniformly complete or finite in the strongest possible sense. Webster’s categorified knot invariants may be infinite-dimensional in general, and full cobordism functoriality is conjectural. This does not weaken the underlying point; rather, it clarifies that the representation-theoretic mechanism fixes the invariant structure, while questions of finiteness, canonical normalization, or full functoriality remain separate technical issues (Webster, 2010).

Taken together, these works suggest that representation-protected invariants form a broad technical paradigm: invariants are encoded in admissible representations of hidden state, symmetry, or categorical data; triviality is the possibility of deforming that representation to a canonical form; and protection is the statement that the representation cannot be altered without violating the relevant typing rule, admissibility condition, symmetry constraint, or categorical equivalence.