Skip to content

Latest commit

 

History

History
303 lines (245 loc) · 14 KB

File metadata and controls

303 lines (245 loc) · 14 KB

Conditional deterministic complete integer factorization

Attribution: icekylinx and GPT-6 Astra Pro.

1. Theorem and accepted external premise

Let N be a positive integer and n=bit_length(N). Bounded inputs are handled directly. Assume the Boolean tensor-contraction theorem identified in sources.md, with its actual succinct-entry and coefficient-ring interface. There is a uniform deterministic classical algorithm returning the complete prime factorization of every input, in

Formula

bit operations and polynomial space in n. No balance, squarefreeness, smoothness, or random-input promise is imposed. The reduction needs at most one nontrivial tensor contraction; powers of two and fully removed small-factor inputs need none. Three resource choices are proved below. The external contraction theorem is an accepted premise, not a conclusion of the local tests or this review.

The literal external interface takes a bipartite Boolean multigraph with O(s) edges, polynomially many vertices, polynomial description length and maximum degree at most ceil(s^(9/10)). Tensor entries are computed uniformly from their ordered Boolean ports. Polynomial-time, polynomial-space local procedures suffice for its allowed entry-computation budget. Its ring is

Formula

for a fixed constant C. It returns the exact contraction in the displayed subexponential time and polynomial space. The reduction checks this specific interface. It does not use a theorem for arbitrary field-valued tensor entries.

2. A unique-extension verifier with linearly many Boolean edges

Remove the power of two and write M for the remaining odd integer. A root witness consists of an even integer x satisfying 0<x<M-1 and

Formula

For every admitted root, the nonnegative quotient y<M is unique. Choose a radix with Theta(log n) bits per digit and pad the digit vectors to a square power-of-two transform length m=Theta(n/log n) exceeding twice their length. Let H=sqrt(m). Linnik's established least-prime bound permits deterministic polynomial-bit preprocessing of a prime pi=n^{O(1)} with a suitable primitive m-th root of unity and magnitude larger than all coefficient/carry bounds. This is a search over polynomially sized integers, not an unproved prime-gap promise. The source-derived root verifier gives the exact constants and signed carry bound.

Compute three grouped forward transforms: the digits of x, the digits of y, and the signed carries. The transform of the known digit vector of M is precomputed. The frequency equations are

Formula

The inverse transform is invertible, so these equations enforce the local carry identities modulo pi. Each integer expression is smaller in absolute value than pi; the congruences are exact integer identities. Their weighted sum telescopes to the original root equation. Canonical field residues, bounded signed carries, and binary comparators remove encoding ambiguity. The carries and all transform words are uniquely determined by x.

Each transform is split into two layers of H-input, H-output gates. A gate has O(H*log n) Boolean ports; its truth predicate is exact polynomial-time field arithmetic. A whole layer uses O(m*log n)=O(n) Boolean edges. The three transforms, pointwise tests, comparators and constant-degree copying therefore use linearly many edges, with maximum degree

Formula

for sufficiently large inputs. Equality subdivision makes the graph bipartite and changes the edge count only by a constant. Truth tables are succinct procedures and are never expanded. Accepting assignments are in bijection with the selected roots; the verifier is constructible without knowing any prime factor. Internal field arithmetic occurs inside Boolean predicates.

3. A separating root-moment family

For analysis only, write M as a product of pairwise coprime odd prime powers q_i=p_i^(e_i), with k distinct primes. At each prime power a square root of one is either plus or minus one. Negation exchanges the parity of representatives modulo odd M. Selecting the even member of each sign pair and removing M-1 therefore leaves the root set R with

Formula

The algorithm does not know the prime powers. A computable bound is provided by the largest K for which (K+1)!<=M: distinct prime factors imply k<=K, and elementary factorial growth gives K=O(n/log n).

Choose m digit blocks, b=ceil(n/m) and radix B=2^b. For each block form an ordinary integer moment and its unknown signed analogues:

Formula

Here epsilon_i(x) is the sign of x modulo q_i. Thus S_j is congruent to A_(i,j) modulo the full prime power.

For distinct components i,h, assume all digit-sum differences vanish modulo p_i. Reassemble the digits and reduce modulo this prime. Exactly half the sign vectors have unequal signs at these two components, and this property is invariant under global sign reversal. The removed trivial sign pair has equal signs. Consequently

Formula

a contradiction. Some digit index separates each ordered pair. Taking c=A_(h,j) shows that a member of the explicit family gcd(M,S_j-c), with |c|<=(B-1)T, contains all of q_h and none of p_i.

Replace every such divisor d by gcd(M,d^n mod M). Since each exponent is at most n, this saturates its prime support: every prime-power component is included in full or not at all. Streaming gcd refinement by these unitary divisors separates every pair, so the final blocks are precisely the prime powers. Exact integer-root tests recover each prime and multiplicity. If T=0, the residual is already a prime power and the same root tests suffice. The number of candidates costs 2^{O(b+K)} time; the list of blocks and loop counters have polynomial size. No prime-factor-size assumption is hidden in the separation lemma.

4. One literal-ring contraction for all moments

For block count m, use truncation ell=2m+1. At the Boolean input bit x_i in block j, attach

Formula

Products of two markers vanish. A two-state chain supplies the weight x+V^m by a single possible transition whose bit weight is 2^i*x_i. The contraction returns

Formula

Taking P=n+b+K+2 prevents wraparound in every coefficient. All required moments and T are read as exact nonnegative integers. Unary markers and the constant-state chain add linearly many edges; no gate must inspect a whole large digit block. With m=ceil(log_2 n), the parameters and recovery work are

Formula

They fit the external theorem with s=Theta(n) and a fixed C. Auxiliary work has smaller exponential order than the contraction. Complete recovery needs one call, rather than adaptive factor-prefix queries or recursive contractions. The earlier multiplication/prefix reduction is retained as an independently useful, weaker call-count construction.

5. Square-root precision from Sidon second moments

First trial-divide through 2^ceil(sqrt(n)), removing every encountered factor to full multiplicity. For the residual of bit length n', its number of prime factors is at most K=ceil(n'/ceil(sqrt(n))). Choose m=ceil(sqrt(n')) blocks and b=ceil(n'/m). Both are of square-root size.

The Sidon construction assigns shifted exponents a_j=O(m^2) with distinct unordered pair sums, no single/pair collision, and all triple products above the truncation degree. The bit weight is the exact quadratic polynomial

Formula

Writing G_x(V)=sum_j d_j(x)V^(a_j), multiplication gives 1+2G_x+2G_x^2 after truncation. The ordinary integer coefficients give twice the diagonal and four times the off-diagonal digit second moments. Precision P=2b+K+3 prevents wraparound; division by two or four happens only after reading the exact integer coefficient. Recombining these second moments gives the previous S_j, hence the same complete separation algorithm. Resources are

Formula

One literal-ring contraction suffices. The leading overall time class is unchanged.

6. Subpolynomial precision and polynomial recovery

The packed square-zero proof uses trial division through an exponent of order n*loglog(n)/log(n). The rough residual has at most K=2^{O(log n/loglog n)}=n^{o(1)} selected roots. Encode roots injectively in an explicitly constructed characteristic-two field of degree r=Theta(n'). A deterministic Artin–Schreier tower constructs it in polynomial time, with no random irreducible-polynomial search.

In the auxiliary square-zero algebra, the elements

Formula

contain the first 2K powers as coordinates. Group the input bits and compile multiplication by these elements into succinct binary regular-representation matrices of dimension D=2rK. Monomial masks only increase along a nonzero path; when a mask stays fixed, the field-coordinate block is the identity. Every integer path has at most q=O(log K) strict transitions. Its counts therefore fit guarded lanes of width W=O(log K*log n).

An output tensor packs Q=ceil(D/n) lanes per polynomial coefficient. With ell=ceil(D/Q)<=n and P=QW, every coefficient is below its modulus and is recovered exactly before reducing its lanes modulo two. The external theorem is still called over its literal integer/truncated-polynomial ring. The auxiliary field algebra defines binary tensor predicates and is not an added coefficient-ring assumption.

The first 2K field power sums determine all distinct roots. The K-by-K Hankel matrix factors as VV^T; the Vandermonde matrix is injective and its transpose is surjective, so its rank is exactly the number of roots, including in characteristic two. The leading square block determines their monic root polynomial. Deterministic trace splitting recovers all linear factors, and gcds with x-1 separate every prime-power component. Postprocessing is polynomial. The resulting resources are

Formula

Formula

Added matrix-state edges are O(n^(1/10)*log D)=o(n) and grouped degree stays below the allowed threshold. Polynomially described truth predicates, linear edge count and legal precision/truncation parameters complete the theorem's applicability check. This option and the Sidon option improve different resources; neither improves the displayed leading exponent.

7. Significance and exact review boundary

The result is a conditional classical deterministic subexponential algorithm in the input bit length. It improves the positive constant exponent in the cited general deterministic factoring bound to a vanishing exponent in the numerical input. It remains superpolynomial in bit length, and the displayed rate is asymptotically slower than the established randomized subexponential factoring bound. The current construction is not a practical performance claim.

The source-derived parameter limitations restrict specified contraction templates. They prove neither optimality of the leading exponent nor a lower bound for general factoring. Separate direct one-fifth-exponent research is retained in the research-branch overview. It is outside this conditional theorem's review.

Independent internal reviews checked the root verifier, both moment extraction methods, literal-ring packing, characteristic-two recovery, actual external interface and resource accounting. The finite checks supplement these arguments. They do not implement or establish the assumed contraction theorem, prove the asymptotic bound mechanically, certify priority, or constitute external peer review or proof-assistant verification.