Hardware Validation of DAGI via a Modular “Ridge” Signature and High-Order Synergistic Information

Petr Sramek · 2026

Read Hardware Validation of DAGI via a Modular “Ridge” Signature and High-Order Synergistic Information by Petr Sramek (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-15051.txt.

Introduction

Sramek reports hardware validation of DAGI on IBM Quantum hardware with a small experiment. Two 4-bit registers with modulus 16 give v congruent to k times u mod 16, forming a ridge in the joint distribution. On ibm_torino in one Sampler V2 job with 8 keys and 1024 shots per key, the ridge persists under noise at 2.93 times baseline. Key recovery beats chance per shot and per group. A Möbius-based decomposition shows targeted positive synergy at order 3, significant under label-shuffle tests, supporting DAGI detection of resilient higher-order structure.

Catalog record

CREATORPetr Sramek
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18101
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.15051

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hardware Validation of DAGI via a Modular “Ridge” Signature and High-Order Synergistic Information Petr Sramek Abstract We report a hardware validation of the DAGI (Directed Acyclic Graph Information) framework on IBM Quantum hardware using a small, controlled experiment whose ideal output distribution is constrained to a low-dimensional modular manifold (a “ridge”). For two n n -bit registers ( u , v ) (u,v) with n = 4 n=4 (modulus 16 16 ), each key instance k k induces an ideal relation v ≡ k ​ u ​ ( mod ​ 16 ) v\equiv k\,u\ (\mathrm{mod}\ 16) , producing a visually distinct ridge in the joint ( u , v ) (u,v) distribution. Executed on ibm_torino in a single SamplerV2 job (8 keys, 1024 shots/key, N = 8192 N=8192 total shots), the ridge persists under hardware noise with ridge-hit probability p hit = 0.1830 p_{\mathrm{hit}}=0.1830 (uniform baseline 1 / 16 1/16 ), corresponding to a ridge contrast of 2.93 × 2.93\times (95% bootstrap CI [ 2.80 , 3.06 ] [2.80,3.06] ). Key recovery exceeds chance: per-shot accuracy 0.1689 0.1689 (chance 0.125 0.125 , 95% Wilson CI [ 0.1610 , 0.1772 ] [0.1610,0.1772] ), and per-group dictionary recovery 0.375 0.375 (chance 0.125 0.125 ). To test the central DAGI hypothesis—that recoverable key information is predominantly high-order/synergistic rather than visible in low-order marginals—we compute a Möbius-based information decomposition of I ⁡ ( K , D S ) \mathrm{I}(K;D_{S}) over detector-bit subsets S S via a Möbius inversion pipeline and evaluate targeted positive synergy CPS K \mathrm{CPS}_{K} at order k max = 3 k_{\max}=3 . We observe CPS K ​ ( k = 3 ) = 0.08788 \mathrm{CPS}_{K}(k=3)=0.08788 with significance under label-shuffle permutation tests (accuracy p = 0.001996 p=0.001996 , CPS K \mathrm{CPS}_{K} p = 0.004975 p=0.004975 ). Uniformity diagnostics show near-uniform single-bit marginals while correlation concentrates in specific low-order pairs, and a bootstrap reliability sweep confirms order-3 targeted synergy remains statistically reliable at the full 1024-shot target budget. These results support the claim that DAGI detects and quantifies nontrivial, hardware-resilient, higher-order information structure associated with a known global algebraic constraint. 1 Introduction Modern quantum devices exhibit complex, correlated noise, and the observables accessible to hardware are often indirect and noisy views of latent algorithmic structure. A recurring practical problem is therefore: given only classical measurement bitstrings from hardware, how can we detect, quantify, and localize meaningful structure that survives noise? The DAGI framework addresses this by combining (i) information-theoretic observables (typically mutual information between a label/latent variable and subsets of measured bits), (ii) Möbius inversion on the subset lattice to decompose information across interaction order, and (iii) reliability controls (e.g., bootstrap-based frontiers) to avoid over-interpreting high-order terms when sampling noise dominates. The DAGI (Directed Acyclic Graph Information) framework used here is developed in a set of core manuscripts that (i) introduce the DAG-based ontology for informational branching and record formation, and (ii) formalize the multiscale Möbius-inversion machinery used to decompose I(K;D_{S}) across subset order and to quantify irreducible (synergistic) structure. The foundational DAG-based interpretation and the Möbius-enhanced multiscale formalism underlying the metrics reported in this paper are described in Refs. [ 5 , 4 ] . The present work should be read as a hardware validation of those definitions and diagnostics on noisy quantum measurement data. In this paper we validate DAGI on a deliberately simple but diagnostically sharp hardware target: a small Fourier/oracle-style circuit family whose ideal output distribution concentrates near a modular manifold (a “ridge”) in the joint outcome space (u,v) . This provides a known global algebraic signature against which we can test two claims: 1. Ridge survival under hardware noise: the low-dimensional ridge remains visible on quantum hardware for small n . 2. High-order key information: information about the key k is not reducible to single-bit marginals, but appears as higher-order (synergistic) structure that DAGI can quantify and validate statistically. 2 Experimental Overview 2.1 Ridge relation and observable variables We measure two n -bit registers (a,b) with n=4 , interpret them (little-endian) as integers u=\mathrm{int}(a)\in\mathbb{Z}_{16},\qquad v=\mathrm{int}(b)\in\mathbb{Z}_{16}, (1) and evaluate a key-dependent ridge residual r_{k}(u,v)\;=\;(v-k\,u)\bmod 2^{n}. (2) The idealized signature for a given key k is concentration of probability mass near the ridge v\equiv k\,u\pmod{16}\quad\Leftrightarrow\quad r_{k}(u,v)=0. (3) We consider a small key set K=\{1,3,5,7,2,4,8,12\}\subset\mathbb{Z}_{16}, (4) chosen to include both odd and even keys (to probe aliasing/coset behavior under modulus 2^{n} while keeping the total hardware cost low). 2.2 Hardware run and dataset All key instances were executed as separate circuits inside a single IBM Runtime SamplerV2 job on the ibm_torino backend: • Backend: ibm_torino • Job ID: d541drnp3tbc73amunv0 • Registers: n=4 (8 measured bits total) • Keys: |K|=8 • Shots per key: 1024 • Total shots: N=8192 • Transpilation: optimization level 1; AQFT disabled Transpiled circuit depths ranged from 334 to 476, with two-qubit gate counts from 188 to 253 (mean depth 411.1; mean 2Q count 224.5). 3 Metrics and Statistical Tests 3.1 Ridge hit and ridge contrast Define the ridge-hit indicator for a given key k : L_{k}(u,v)=\mathbf{1}[r_{k}(u,v)=0]. (5) Let p_{\mathrm{hit}} denote the empirical ridge-hit probability under the true key label. Under a null model where (u,v) is uniform over \mathbb{Z}_{16}\times\mathbb{Z}_{16} , the ridge-hit probability is exactly 2^{-n}=1/16 . We define ridge contrast as \mathrm{Contrast}\;=\;\frac{p_{\mathrm{hit}}}{2^{-n}}. (6) 3.2 Key recovery metrics We report two complementary key-recovery metrics: 1. Per-shot key accuracy: a classifier predicts \hat{k}(u,v) from each individual outcome, and accuracy is \Pr[\hat{k}=k] . 2. Per-group dictionary recovery: for each true key k , aggregate its 1024 shots and select the key that maximizes ridge-hits (or minimizes ridge distance) across that group. This corresponds to a simple “dictionary attack” style recovery and is reported as group accuracy over the 8 key groups. Chance accuracy for both metrics is 1/|K|=0.125 . 3.3 DAGI / Möbius information decomposition Let D=(D_{1},\dots,D_{2n}) denote the measured bits (here 2n=8 ). Let K denote the key label (discrete, 8 classes). For any subset of measured bits S\subseteq\{1,\dots,2n\} , define g(S)\;=\;\mathrm{I}(K;D_{S}), (7) the mutual information between the key label and the subset of measured bits. We compute a Möbius inversion on the subset lattice to obtain a decomposition g(S)\;=\;\sum_{T\subseteq S}f(T), (8) where f(T) can be interpreted as the irreducible (non-overlapping) contribution attributable to subset T . Positive f(T) at |T|\geq 3 indicates synergistic information about K that requires jointly observing the bits in T . We summarize by order k via the positive mass : M_{k}^{+}\;=\;\sum_{|T|=k}\max\{f(T),0\}. (9) We report the targeted order-3 positive synergy score \mathrm{CPS}_{K}(k=3)\;\equiv\;M_{3}^{+}, (10) computed with k_{\max}=3 . We further partition subsets by whether they lie within a register or are cross-register (touching at least one bit from each of a and b ), and report within/cross decompositions of positive mass. 3.4 Reliability frontier To ensure higher-order estimates are not artifacts of finite sampling, we evaluate a bootstrap/subsampling reliability sweep over shots per key. For each shots/key budget, we estimate a coefficient of variation \mathrm{CV}_{k} for M_{k}^{+} (or its bootstrap estimate) and define the max reliable order k^{\star} as the largest k satisfying \mathrm{CV}_{k}\leq 1. (11) 3.5 Permutation tests We test whether the observed key accuracy and \mathrm{CPS}_{K} could arise under a null model by shuffling key labels across shots (label-shuffle permutation). The empirical p -value is computed as the fraction of permutations with

Read the complete text