Amplification-by-Independence Argument

• Amplification-by-independence argument is a framework where weak independence or irrelevance conditions lead to forced precision, factorization, and mixing in models.
• It spans disciplines from decision theory to quantum information and non-Hermitian physics, transforming imprecise or path-dependent quantities into robust outcomes.
• This argument unifies various fields by showing that introducing symmetry or independence conditions catalyzes structural collapse and enhances randomness extraction.

The amplification-by-independence argument encompasses a suite of rigorous results in decision theory, probability, computational randomness, quantum information, and non-Hermitian physics, where independence or irrelevance assumptions yield amplification, factorization, or purification effects in the structure of models, extracted randomness, or physical observables. The technical core of the argument is that introducing certain symmetries, independence, or irrelevance relations can have disproportionately strong or even “collapsing” consequences for the underlying structure, often forcing strong mixing, singling out admissible rules, or rendering path-dependent quantities path-independent. This phenomenon plays a central role in domains such as imprecise probability models, device-independent randomness amplification, Kolmogorov complexity, and geometric amplification in non-Hermitian systems. The following sections organize the foundational results and applications of the amplification-by-independence argument across these contexts.

1. Seidenfeld’s Irrelevance, Independence, and E-Admissibility

Teddy Seidenfeld’s critique of classical binary preference models in decision theory led to the formulation of irrelevance (S-irrelevance) and independence (S-independence) judgments in the language of sets of desirable option sets (Bock et al., 2021). Let Ω be a possibility space, and consider a set K of desirable option sets (collections of bounded real-valued gambles). The key technical condition is that event E is S-irrelevant to F with respect to K (denoted $E \rightarrow_s F$) if, for every F-gamble f and every ε > 0, the composite gamble set $$A_\epsilon = \{f, g, 1_E \cdot f + 1_{E^c} \cdot g - \epsilon \}$$ (g arbitrary on F) never lets one pay ε to exploit E for improved decision making. This precise formulation forces strong structural constraints.

In the case where K arises from a linear prevision P, S-irrelevance between events E and F is equivalent to probabilistic independence: $P(E \cap F) = P(E) P(F)$. When coherent lower previsions (interval-valued models) are used, even a single S-irrelevance assessment magnifies to a collapse: unless E or F is trivial, the lower prevision becomes precise on F (admits a unique dominating linear prevision), and factorization follows: $\mathcal{L}(h \cdot g) = \mathcal{L}(h) P_F(g)$ for disjoint-event functions h and g.

2. Collapse to Mixing and E-Admissibility

Extending to non-binary and non-convex models, the amplification-by-independence argument demonstrates that as soon as one admits an Archimedean set of desirable option sets $$K = \cap_{\mathcal{L} \in \mathcal{M}} K_{\mathcal{L}}$$ and asserts S-irrelevance between credibly indeterminate events E and F, all $\mathcal{L} \in \mathcal{M}$ become precise on F. Consequently, the marginal model K[F] is forced to be mixing—i.e., representable as an intersection of K_P for a set of linear previsions—so that the choice function on F-gambles reduces to the union-of-maximizers (“E-admissibility”): $$\text{choose}(A) = \bigcup_{P \in P(\mathcal{M})} \arg\max_{f \in A} P(f)$$.

This amplification thus forces all uncertainty on F into mixtures over precise models, regardless of initial imprecision or non-Archimedean structure, so long as weak irrelevance is upheld. E-admissibility, a rule traditionally justified by universal maximality, emerges as the unique survivor of honoring weak irrelevance, providing a compelling structural motivation for its adoption (Bock et al., 2021).

3. Amplification-by-Independence in Randomness and Extractors

A related form of amplification-by-independence appears in computational complexity and randomness extraction. In Kolmogorov complexity theory, if two n-bit strings x and y have dependency dep(x,y) at most α(n), computable extractors can output a string z of length roughly s(n) with conditional complexity ≥ s(n) – α(n) given either input. However, there is a sharp impossibility result: no pair of computable functions f₁, f₂ can reduce the dependency below α(n) – O(log n), i.e., no independence amplification beyond logarithmic slack is achievable (Zimand, 2010).

Core implication: The budget of independence present in the inputs fully constrains what randomness or “purified” independence can be extracted—one cannot engineer higher independence via deterministic transformation, reflecting the “no-free-lunch” restriction. This impossibility provides a technical analogue to the collapse phenomena in imprecise probability, where irrelevance and independence assessments can only eliminate uncertainty, not create more independence than given initially.

4. Device-Independent Randomness Amplification via Independence

Device-independent randomness amplification protocols further illustrate amplification-by-independence, leveraging no-signaling and input-source independence rather than detailed device models (Ramanathan et al., 2015, Kessler et al., 2017). In these protocols, weak random sources (e.g., Santha–Vazirani sources) are used to select measurement settings for a two-component device subject only to no-signaling constraints. A Bell test provides partial certification of randomness (randomness in a single input-output pair), and a partial tomography step ensures that for a positive fraction of trials, no non-signaling box can output deterministically.

The key amplification-by-independence phenomenon is that the protocol’s security reduces to the independence between the weak random source and the device (and side information). Given this minimal assumption, and via extractors designed for the Markov model, the workflow amplifies the original minuscule independence into full cryptographic-quality randomness, provably secure even against general no-signaling adversaries. This effect—amplification of min-entropy and privacy entirely from independence—demonstrates the foundational leverage of independence assumptions in the geometry of security proofs.

5. Geometric Amplification and Path-Independence in Non-Hermitian Systems

In non-Hermitian quantum and classical systems, the amplification-by-independence effect arises as geometric amplification of state norms under adiabatic evolution (Ozawa et al., 2024). The non-Hermitian Berry phase generically has an imaginary part, giving rise to “geometric gain” (or decay). When the imaginary part of the Berry curvature vanishes, the total geometric amplification is path-independent—determined solely by the initial and final points in parameter space—thus establishing an amplification-by-independence effect.

This path independence is guaranteed when left and right eigenstates of the Hamiltonian are related by symmetries (unitary or projector), and the geometric amplification factor reduces to an explicit ratio of Petermann factors at the endpoints. For example, in unitarily symmetric cases, $G_{\mathrm{geo}} = \sqrt{K_f/K_i}$; for rank-1 projectors, $G_{\mathrm{geo}} = K_f/K_i$, where K is the Petermann factor quantifying non-orthogonality. Experimental demonstrations in both non-Hermitian two-level systems and classical metamaterials confirm the path independence and endpoint formulae, connecting the phenomenon rigorously to structural symmetries and their consequences for measurable amplification.

6. Synthesis: Structural Consequences and Unifying Principles

The amplification-by-independence argument consistently exhibits that weak independence or irrelevance conditions—whether in probabilistic, algebraic, computational, or geometric contexts—lead to forced precision, mixing, path independence, or collapse of structure. In all cases, the introduction of independence induces highly nontrivial constraints on the system, often compelling selection rules, factorization, or maximal extractable randomness.

A unified theme is that independence and irrelevance assumptions serve as catalytic conditions, amplifying modest structural properties into strong global effects (collapse to E-admissibility, full randomness, or geometric gain), and typically remove model or protocol degrees of freedom. This identifies amplification-by-independence as a guiding structural principle across mathematical physics, decision theory, and theoretical computer science.