Orbit-Collapse Complexity: A Symmetry-Based Lens on P, NP, and Structured Cores

1. Introduction

The P vs NP question asks whether every language whose YES-certificates can be verified in polynomial time also admits polynomial-time decision procedures. This work develops a symmetry- and type-theoretic lens, Orbit-Collapse Complexity (OCC), that (i) proves sufficiency criteria for polynomial-time decidability, (ii) quantifies the structured mass of tractable subfamilies, and (iii) provides reduction policies that target hardness outside those subfamilies. Our contributions are:

• Collapse ⇒ P (Theorem 1): If instances canonically compress to polynomially many types with polytime canonicalization and per-type solvers, the language is in P.
• Metrics for structure: orbit coverage, type-growth exponents, and orbit-compressibility index make “how much collapses” a measurable notion.
• Distance-to-core algorithms: XP and FPT meta-theorems for instances at bounded edit distance from a P-core.
• Layer-Respecting Reductions (LRR): reductions that preserve or raise layer depth, supporting hardness beyond cores.
• Sudoku case study: exact separable class law, bi-affine degree-4 certificates, and Pólya formulas for piecewise families.

2. Model and Symmetries

We consider finite-domain CSP families; the discussion covers SAT, Graph Coloring, Sudoku/Latin, Integer Programming encodings, TSP with discrete symmetries, etc.

Definition 1 

(CSP family). A CSP family $\mathcal{L}={\left\{L_n\right\}}_{n\ge 1}$ has instances of size n over a finite domain D. Each instance $I\in L_n$ specifies constraints of bounded arity on variables $V\left(I\right)$ with $\left|V\right(I\left)\right|=\mathrm{poly}\left(n\right)$. The language is the decision problem: “Does I admit a solution?”

Definition 2 

(Symmetry group). For each n, let $G_n$ be a finite group that acts on instances and solutions. The orbit of X is $G_n·X=\{g·X:g\in G_n\}$ and the stabilizer is ${\mathrm{Stab}}_{G_n}\left(X\right)=\{g:g·X=X\}$.

Definition 3 

(Orbit-compressibility). For a solution S, define $\kappa \left(S\right)=1-{\displaystyle \frac{log|{\mathrm{Stab}}_{G_n}\left(S\right)|}{log|G_n|}}\in [0,1].$ Small κ indicates large stabilizer (high structure).

3. Constructors, Layers, and P-Cores

Definition 4 

(Constructor, Layer). Aconstructor $\mathcal{C}$ is a uniform polytime recognizer that defines a syntactic subfamily $K_n^{\mathcal{C}}\subseteq L_n$. A list ${\mathsf{S}}_0,{\mathsf{S}}_1,\dots$ defines layers $S_t\left(n\right)=K_n^{{\mathsf{S}}_t}$. A fixed layer order assigns each solution to the first matching layer; layers become disjoint by fiat.

Definition 5 

(P-core). A P-core is a constructor family $K=\left\{K_n\right\}$ with polytime membership (and, when needed, polytime canonicalization within its symmetry). Trivial finite sets are excluded.

Proposition 1. 

If membership in $K_n$ is decidable in time $\mathrm{poly}\left(n\right)$, then deciding L restricted to K is in P.

4. Orbit-Collapse ⇒ P

Definition 6 

(Types and canonicalization). Fix an equivalence ∼ (e.g., the $G_n$-action or a coarser invariant). Let ${\mathcal{T}}_n$ be the set of canonical representatives. A canonicalizer is a polytime map $\mathrm{can}\left(I\right)\in {\mathcal{T}}_n$ with $I\sim \mathrm{can}\left(I\right)$.

Theorem 1 

(Orbit-Collapse ⇒ P). Let $\mathcal{L}=\left\{L_n\right\}$. Suppose for all n: (i) $|{\mathcal{T}}_n|\le n^c$; (ii) $\mathrm{can}\left(I\right)$ is computable in $\mathrm{poly}\left(n\right)$; (iii) there is a $\mathrm{poly}\left(n\right)$ predicate A on ${\mathcal{T}}_n$ with $I\in L_n\iff A\left(\mathrm{can}\left(I\right)\right)=1$. Then $\mathcal{L}\in \mathrm{P}$.

Proof. 

On input I, compute $T=\mathrm{can}\left(I\right)$ and return $A\left(T\right)$. Correctness is by (iii). Runtime is polynomial by (ii)–(iii). The bound (i) ensures that the analysis of types is polynomially bounded and realizable.    □

5. Quantifying Collapse

Let ${\mathcal{O}}_n\left(L\right)$ be solution orbits in $L_n$ under $G_n$; ${\mathcal{O}}_n\left(K\right)$ for a core K. We use:

• Orbit coverage ${\mu }_n(K\mid L)={\displaystyle \frac{|{\mathcal{O}}_n\left(K\right)|}{|{\mathcal{O}}_n\left(L\right)|}}$;
• Type-growth exponent $\gamma \left(K\right)={lim\; sup}_{n\to \infty }{\displaystyle \frac{log|{\mathcal{T}}_n\left(K\right)|}{logn}}$;
• Core ratio inside a layer ${\rho }_{K\mid R}\left(n\right)={\displaystyle \frac{|{\mathcal{O}}_n\left(K\right)|}{|{\mathcal{O}}_n\left(R\right)|}}$ for $K\subseteq R\subseteq L$.

6. Algorithms by Distance-to-Core

Definition 7 

(Edit distance). Fix a finite set Δ of local edit templates. For instance I, let ${\mathrm{dist}}_{\Delta }(I,K)$ be the minimum number of Δ-edits sending I into K.

Lemma 1 

(XP by bounded edits). For bounded-arity CSPs and any P-core K, satisfiability is decidable in time $n^{O\left(d\right)}·\mathrm{poly}\left(n\right)$, where $d={\mathrm{dist}}_{\Delta }(I,K)$.

Proof. 

Branch over $n^{O\left(d\right)}$ candidate edit sets of size d, test membership in K in polynomial time, and verify consistency.    □

Theorem 2 

(FPT under bounded boundary width). Assume: (i) constraints have bounded arity; (ii) any set of d edits induces a defect substructure of treewidth $w\left(d\right)$; (iii) the core K is MSO-definable on bounded-treewidth hosts. Then satisfiability is decidable in time $f\left(d\right)·\mathrm{poly}\left(n\right)$.

Proof sketch. 

Enumerate the interface between the edited region and K; apply DP/Courcelle on a tree decomposition of width $w\left(d\right)$ to glue feasible assignments consistent with K. All combinatorics depend on d only.    □

7. Layer-Respecting Reductions (LRR)

Definition 8 

(Layer depth). Given layers $S_0,S_1,\dots$, define $\ell \left(I\right)=min\{t:I\mathrm{has}\mathrm{a}\mathrm{solution}\mathrm{in}{S}_{\le t}\}$.

Definition 9 

(LRR at depth t). A reduction $R:A\to B$ is layer-respecting at depth t if $\ell \left(R\right(I\left)\right)\ge min\{\ell (I),t\}$ for all I.

Proposition 2 

(Closure). If $R_1$ is LRR at depth $t_1$ and $R_2$ at depth $t_2$, then $R_2\circ R_1$ is LRR at depth $min\{t_1,t_2\}$.

Proof. 

Immediate from the definition.    □

Conjecture 1 (Layered ETH (L-ETH)). 

For each fixed t, there exists$c_t>0$ such that deciding instances with $\ell \left(I\right)>t$requires time$\ge 2^{c_{t}n}$(under ETH/SETH-style assumptions), witnessed by LRRs from 3-SAT that pushℓabove t.

8. Layered Orbit Sums (Accounting)

Let ${\mathsf{\Omega }}_n$ be all solutions at size n, with layer-ordered assignment (each solution taken by the first matching layer). Then:

Theorem 3 

(Layered Orbit Sum). $|{\mathsf{\Omega }}_n|={\sum }_L{\sum }_{\left[S\right]\in L/G_n}{\displaystyle \frac{|G_n|}{|{\mathrm{Stab}}_{G_n}\left(S\right)|}},$ with an extra factor $m!/|{\mathrm{Aut}}_{\mathrm{s}ym}\left(S\right)|$ if symbol relabeling is counted, where $m=\left|D\right|$.

Proof. 

Orbit–stabilizer plus disjointness of layer assignment.    □

9. Proof-Complexity Calibration inside Layers

Proposition 3 

(Bi-affine ⇒ low-degree certificates). In bi-affine layers (e.g. Sudoku $S_1$), feasibility/infeasibility has polynomial-size polynomial-calculus and degree-4 sum-of-squares certificates after standard symmetry breaking.

Proof sketch. 

Encode row/column/box bijectivity by quadratic equations over $0/1$ indicators; the bi-affine placement map linearizes under a fixed normal form so constraints reduce to degree $\le 2$ identities. SOS/PC derives these with degree $\le 4$.    □

10. Sudoku Case Study: Formulas and Laws

Let k be the box order; grid size $n=k^2$; position symmetry $G_k={(S_k\wr S_k)}_{\mathrm{r}ows}\times {(S_k\wr S_k)}_{\mathrm{c}ols}$ with canonical digit relabeling.

Separable layer $S_0$

Two orbit sizes arise: large $=2\phi {\left(k\right)}^2$ and small$=\phi {\left(k\right)}^2$. Let $r_2\left(k\right)=\left|\{x\in {\left({\mathbb{Z}}_k\right)}^{\times }:x^2\equiv 1(modk)\}\right|$, $v_2\left(k\right)$ the 2-adic valuation, and ${\omega }_{\mathrm{o}dd}\left(k\right)$ the number of distinct odd primes dividing k.

Theorem 4 

(Separable class law). $S_{\mathrm{s}ep}\left(k\right)={\displaystyle \frac{\phi {\left(k\right)}^2+E\left(k\right)}{2}}$ with $E\left(k\right)=\left\{\begin{array}{cc}2r_2\left(k\right),\hfill & v_2\left(k\right)\le 1\mathrm{and}{\omega }_{\mathrm{o}dd}\left(k\right)\le 1,\hfill \\ 4r_2\left(k\right),\hfill & \mathrm{otherwise}.\hfill \end{array}\right.$ Moreover #small classes $=E\left(k\right)$ and #large $=(\phi {\left(k\right)}^2-E\left(k\right))/2$.

Proof outline. 

Parameterize separable constructors by $(\alpha ,\beta )\in {\left({\left({\mathbb{Z}}_k\right)}^{\times }\right)}^2$ producing type pairs $(s,t)$. The group $G_k$ together with internal permutations on $(s,t)$ acts on this parameter space. By the Chinese Remainder Theorem, ${\left({\mathbb{Z}}_k\right)}^{\times }\simeq {\prod }_{p^e\Vert k}{\left({\mathbb{Z}}_{p^e}\right)}^{\times }$ and each factor contributes fixed points corresponding to involutions ($x^2\equiv 1$). These yield index-2 stabilizers responsible for small orbits; counting across CRT components gives a factor $r_2\left(k\right)$. The doubling $E\left(k\right)=2r_2\left(k\right)$ in the prime-lean regime ($v_2\le 1$, at most one odd prime) and $E\left(k\right)=4r_2\left(k\right)$ in composite-rich cases reflects whether both axes admit independent involutive symmetries after normal forms. A detailed orbit–stabilizer enumeration (omitted here for brevity) completes the count.    □

Bi-affine layer $S_1$

Empirically (and provably for small k): $C_{\mathrm{l}in}\left(k\right)=3k^2-5k+\epsilon \left(k\right),$ with $\epsilon \left(k\right)=2$ if $6\mid k$, else 0; and $C_{\mathrm{n}onsep}=C_{\mathrm{l}in}-S_{\mathrm{s}ep}$. For prime $k>3$, $\phi \left(k\right)=k-1$ and the ratio ${\displaystyle \frac{S_{\mathrm{s}ep}}{C_{\mathrm{l}in}}}={\displaystyle \frac{{(k-1)}^2+4}{2(3k^2-5k)}}={\displaystyle \frac{1}{6}}+O(1/k)$ shows a constant-order core inside $S_1$.

Piecewise Layers $S_2,S_3$

With multiplicative local catalogs $T_{*}\left(k\right)={\prod }_{p^e\Vert k}{t}_{*}\left(p^e\right)$ (tabulated by small fixed-point counts), exact Pólya forms are $C_{\mathrm{S}2}\left(k\right)=\frac{1}{2}\left({\displaystyle \genfrac{}{}{0pt}{}{T_U\left(k\right)+k-1}{k}}\right)\left({\displaystyle \genfrac{}{}{0pt}{}{T_V\left(k\right)+k-1}{k}}\right),C_{\mathrm{S}3}\left(k\right)=\frac{1}{2}{\left({\displaystyle \genfrac{}{}{0pt}{}{T_Q\left(k\right)+k-1}{k}}\right)}^2.$

11. Taxonomy by Collapse

• S-OC: complete constructor + polytime canonicalization + $|{\mathcal{T}}_n|\le n^{O\left(1\right)}$⇒ language in P (Thm. 1).
• W-OC: large P-cores with subexponential type growth and nontrivial orbit coverage; full language may remain NP-hard.
• NC: no such collapse under the chosen symmetries; expect NP-hardness beyond cores; target LRR.

12. How to Apply OCC

Pick $G_n$; design constructors ${\mathsf{S}}_0,{\mathsf{S}}_1,\dots$; count types; establish canonicalization and per-type tests; compute ${\mu }_n,\gamma ,\rho$; design distance-to-core algorithms; and craft LRR gadgets for hardness outside cores.

13. What is Proved vs Conjectured

Proved here: Collapse ⇒ P (Thm. 1); layered orbit sum; core-in-P; LRR closure; XP/FPT meta-results (with stated locality assumptions); bi-affine degree-4 certificates (Prop. 3) with constructive sketch.

Conjectural: L-ETH; full proof of the bi-affine count $C_{\mathrm{l}in}\left(k\right)$ for all k; Wreath-invariant SoS lower bounds beyond cores.