CDP: Cyclic Digit-sum Projection — Structural Analysis of SHA-256 Output Distribution and Ergodic Basin Pressure

We introduce CDP (Cyclic Digit-sum Projection), a structural analysis framework for SHA-256 that reveals previously undocumented mathematical properties of the hash function's output distribution. The core observation is that the hex-digit sum W(H) of any SHA-256 output H, when iteratively re-hashed through f(w) = W(SHA256(str(w))), converges deterministically into exactly two closed cycles. This cyclic structure, combined with a multi-component fingerprint, yields a bijective mapping over constrained input spaces enabling O(1) preimage lookup.

Beyond the lookup application, we prove four theorems about SHA-256's internal structure: (1) a complement nibble sum invariant (sum = 38 for all complement byte pairs), (2) convergence of a 20% universal M-rate constant, (3) a universal collapse rule at SHA-256 padding word boundaries, and (4) an ergodic Markov property of SHA-256's compression function under the CDP projection. The Markov chain has transition matrix P = [[0.1812, 0.8188], [0.1677, 0.8323]] with stationary distribution pi_B = 0.1700, independently of round constants K[i], initial values H0, and input class. We further identify a mathematical connection between SHA-256's output structure and AES GF(2^8) arithmetic via the LFSR xtime operation.

Version 2 additions: Extended basin topology analysis reveals the true C1 basin of attraction spans 84 values (16.8% of domain) against C2's 416 values (83.2%), with maximum convergence depths of 8 and 16 respectively. A new structural finding: W(H0) = 502, where H0 denotes SHA-256's NIST initialization constants, is +22.2 units above the equilibrium mean (479.8), creating a measurable round-0 basin-membership deficit of 16.9% — a detectable structural signature of SHA-256's initialization. The universal M-rate theorem, Markov ergodicity claim, and sequence asymmetry observation are also revised with corrected parameters. The C1-basin definition is clarified (25-value core chain vs. 14 direct 1-step attractors vs. 84-value full basin). The convergence bound is corrected to ≤16 iterations. CDP rainbow chain merge behavior is empirically quantified: 1.81× improvement over random reduction, with 66.7% unique chains at saturation.

None of these properties appear in the existing cryptographic literature. CDP does not break SHA-256's preimage or collision resistance; it reveals structural properties of the output distribution under a novel projection.