Quantum Inferential Algebra
docs/Research/PRR-2026-009-quantum-inferential-algebra-notes.md for the planning
record and the in-character derivation session it formalizes.
Abstract
PRR-2026-003 built C, the algebra of concrete invocable operators, on classical Markov kernels, and noted only in passing that this made R ⊂ C resemble a classical/quantum split — an analogy never made to do any actual work. This paper makes it literal. We define Q, quantum-invocable operators whose invocation resolves a prepared Hilbert-space state via the Born rule, and show R ⊂ C ⊂ Q is a genuine tower of embeddings, each level a zero- coherence special case of the next. We prove that any quantum kernel whose measurement commutes with a shared reference basis reduces, in that basis, to an ordinary member of C — genuine quantum behavior requires coherence somewhere composition touches, not just a Hilbert space to live in. We then define coherent pooling, a new combinator native to Q with no counterpart in R or C, and prove — with an explicit, hand-verified two-outcome witness — that it obeys no analogue of PRR-2026-008's Aggregate Pooling Bound: the same two maximally uninformative inputs can be combined, by relative phase alone, into either certain success or certain failure, strictly outside both inputs' own scores in either direction. We give the exact reason the earlier bound's proof does not reach this combinator, extend PRR-2026-008's operator-level correctness score to Q directly, show sequential quantum composition reduces to PRR-2026-003's classical composition once every stage is measured, and report one further correction: PRR-2026-003's own "Unifying Theorem," that invertibility never gets richer under non-determinism, does not survive to the amplitude level.
Keywords. quantum kernels; Born rule; decoherence; coherent superposition; interference; unitary groups; operator-level correctness; Edwardian algebra
1. Introduction
Every operator class this series has built so far — R, C, and the abstract nondeterministic relations A of PRR-2026-003 — varies which outcome an invocation produces. None of them varies the process that produces the outcome itself: a Markov kernel samples from an already-fixed distribution, with no notion of two candidate preparations combining before either is resolved. This paper asks what changes once that assumption is dropped — when the source of an invocation's non-determinism is a Hilbert-space state subject to the Born rule, rather than a distribution already committed to its weights. The answer is not "everything breaks." Most of what this series has proven survives intact, recovered exactly the moment every stage of a quantum process is actually measured. What does not survive is stated precisely, with the exact point of the earlier proof that no longer applies — in keeping with this series' practice of reporting what didn't work alongside what did (PRR-2026-003's unresolved iff-question, PRR-2026-008's rejected Reading B).
2. Preliminaries
We reuse S (the Lindenbaum–Tarski Boolean algebra of PRR-2026-002 §2, countably infinite and atomless), its order ≤, R and its embedding e : R → C, C's finite-support Markov-kernel discipline (PRR-2026-003 §0, Def. 3.1), the correctness coordinate c : S → ℝ with order-compatibility (PRR-2026-005 Def. 3.1–3.2), and PRR-2026-008's operator-level score Corr_{c,μ} (Def. 5.1) together with its Aggregate Pooling Bound (Theorem 8.1): for any κ₁, κ₂ ∈ C, Corr_{c,μ}(κ₁⊞κ₂) ≥ Corr_{c,μ}(κ₁) and ≥ Corr_{c,μ}(κ₂), unconditionally. All are unchanged here.
3. Quantum-Invocable Operators
A quantum kernel is a map q : S → 𝒬(S), q(x) = (H_x, |ψ_x⟩, {Π_y^x}_{y∈F_x}), where H_x is a finite-dimensional complex Hilbert space, |ψ_x⟩ ∈ H_x a unit vector, F_x ⊂ S finite (the same finite-support discipline as PRR-2026-003 §0 — a real invocation still resolves to only finitely many candidates), and {Π_y^x}_{y∈F_x} a projective measurement on H_x: pairwise-orthogonal projectors summing to id_{H_x}.
D : Q → C, D(q)(x)(y) = ⟨ψ_x|Π_y^x|ψ_x⟩ (the Born rule). e_Q : C → Q: for κ ∈ C, fix an orthonormal basis {|e_y⟩}_{y∈F_x} of ℂ^{F_x} and set e_Q(κ)(x) = (ℂ^{F_x}, Σ_y √κ(x)(y)|e_y⟩, {|e_y⟩⟨e_y|}_{y∈F_x}).
D ∘ e_Q = id_C.
Proof.Direct computation: ⟨e_y|(Σ_z√κ(x)(z)|e_z⟩)⟩ = √κ(x)(y), so |⟨e_y|ψ_x⟩|² = κ(x)(y) exactly. ☐
Composing with PRR-2026-003's e : R → C (a deterministic operator embeds as the pure basis state |e_{ρ(x)}⟩ with the trivial one-outcome measurement) gives a literal tower R ⊂ C ⊂ Q, each inclusion a zero-variance/zero-coherence special case of the next — no longer PRR-2026-003 §8/§12's expository-only metaphor. Not claimed: that e_Q is unique, or that every quantum kernel arises this way. It is a section of D, not a characterization of Q's image — most quantum kernels are not diagonal in any basis at all, which is exactly §4's subject.
4. Classical Reducibility of Decoherent Kernels
Within a single measurement, a projective measurement's own outcomes are already pairwise orthogonal by definition — one measurement alone can never exhibit interference between its own outcomes. The question this section answers is different: given two quantum kernels compared against a shared reference basis, when is the comparison itself indistinguishable from ordinary classical randomness?
Fix a shared orthonormal basis {|e_y⟩}_{y∈F_x} at each x (the same one D and e_Q already use for comparison). Call q decoherent at x if its projectors {Π_y^x} pairwise commute in that basis.
If q is decoherent at every x, then q ≅ e_Q(D(q)) up to a change of orthonormal basis.
Proof.A finite family of pairwise-commuting orthogonal projections summing to the identity is simultaneously diagonalizable (spectral theorem for commuting normal operators). In the common eigenbasis {|e_y⟩}, Π_y^x = |e_y⟩⟨e_y| (or a sum over a degenerate eigenspace, which only refines F_x, without affecting the argument). Writing |ψ_x⟩ = Σ_y c_y|e_y⟩ in that basis, the Born rule gives q̂(x)(y) = |c_y|² = κ(x)(y) for κ = D(q) — exactly e_Q(κ)'s own construction, in that basis. ☐
Explicit scope boundary. This proves only that commuting projectors are classical in disguise — an arithmetic fact about a shared comparison basis. It does not attempt the harder, related claim that any measurement scheme respecting S's own ∨/∧ lattice structure must commute — a well-known fact in the physics literature (quantum logic is famously not Boolean; Birkhoff & von Neumann, 1936), cited here as motivating context rather than reproved as this paper's own result. Genuine quantum behavior, this section shows, requires non-commuting structure to enter somewhere a comparison is made — never within one measurement's own outcomes alone.
5. Coherent Pooling and the Limits of Classical Pooling
q₁, q₂ share the same H_x and measurement basis {|e_y⟩}_{y∈F_x}, differing only in prepared state — a genuine, disclosed modeling choice, the same discipline PRR-2026-008 used in flagging μ as a free parameter rather than hiding it.
|ψ_x^{12}⟩ = N(θ)·(|ψ_x^(1)⟩ + e^{iθ}|ψ_x^(2)⟩) N(θ)^{-2} = 2 +
2·Re(e^{iθ}⟨ψ_x^(1)|ψ_x^(2)⟩)
undefined where the two states cancel outright — a flagged domain restriction, disclosed rather than hidden, matching this series' habit of naming edge cases explicitly (cf. PRR-2026-008 Def. 6.1's undefined-cosine edge case).
Worked Example — the Interference Witness
Two exclusive outcomes a, b with c(a) = 100, c(b) = −100.
|ψ^(1)⟩ = (|a⟩+|b⟩)/√2 |ψ^(2)⟩ = (|a⟩−|b⟩)/√2
Individually
Born rule on |ψ^(1)⟩: P(a)=P(b)=½. Same on |ψ^(2)⟩. Both are the least informative state this domain has — a fair coin, by two different routes. Corr_Q(q₁) = Corr_Q(q₂) = ½(100) + ½(−100) = 0.
Combined at θ = 0
|ψ^(1)⟩+|ψ^(2)⟩ = √2|a⟩, normalizes to |a⟩ exactly. P(a)=1, P(b)=0. Corr_Q = 100 — strictly above both inputs.
Combined at θ = π
|ψ^(1)⟩−|ψ^(2)⟩ = √2|b⟩, normalizes to |b⟩ exactly. P(a)=0, P(b)=1. Corr_Q = −100 — strictly below both inputs. Two maximally uninformative states, combined, become the domain's single worst possible outcome, by relative phase alone.
There exist q₁, q₂ ∈ Q and phases θ such that Corr_Q(q₁⊕_Q^θ q₂) < min(Corr_Q(q₁), Corr_Q(q₂)) and, for a different θ, Corr_Q(q₁⊕_Q^θ q₂) > max(Corr_Q(q₁), Corr_Q(q₂)) — the Worked Example above witnesses both, by hand, exactly. Contrast PRR-2026-008 Theorem 8.1: Corr_{c,μ}(κ₁⊞κ₂) ≥ both inputs, unconditionally, for every κ₁,κ₂ ∈ C.
This is not a counterexample to Theorem 8.1 — it is a demonstration that ⊕_Q^θ was never an instance of ⊞ to begin with, and the reason is the actual content of this section, not a footnote. Theorem 8.1's proof runs entirely on a∨b ≥ a, pointwise, for two outcomes that have already been independently sampled; order-compatibility then turns that inequality in S into one in correctness. ⊕_Q^θ never realizes two samples before combining them — it adds un-collapsed amplitudes for two entire preparations, and the Born-rule probability of any single outcome is a quadratic functional of the combined amplitude, not an order-theoretic join of two prior draws. There is no operative instance of "a∨b" anywhere in ⊕_Q^θ's construction for Theorem 8.1's proof to reach.
⊕_Q^θ is, instead, a genuinely new combinator native to Q, with no operation in R or C mapping onto it: unbounded by either input in either direction, its behavior governed entirely by a relative phase with no classical counterpart at all.
6. Quantum Correctness Functional and Sequential Composition
Ĉ_x = Σ_{y∈F_x} c(y)·|e_y⟩⟨e_y| Corr_Q(q)(x) = ⟨ψ_x|Ĉ_x|ψ_x⟩
Corr_Q(q)(x) = 𝔼_{y∼D(q)(x)}[c(y)] — the quantum functional agrees exactly with PRR-2026-008's Corr_{c,δ_x} applied to the decohered kernel.
Proof.Expand Ĉ_x in the Born-rule basis: Σ_y c(y)|⟨e_y|ψ_x⟩|² = Σ_y c(y)·q̂(x)(y), the definition of the classical expectation. ☐
Extended over the whole domain, Corr_{Q,μ}(q) = Σ_x μ(x)·Corr_Q(q)(x), matching PRR-2026-008 Definition 5.1 the instant the state is decohered — nothing in §5's interference result is smuggled in through a different score; it is the same functional, evaluated one layer earlier, before the state has been read off.
q₂ ⊛_Q q₁: the outcome of q₁ is measured and recorded, then q₂ is prepared and invoked from the resulting post-measurement state — the standard quantum-instrument picture, not a new axiom.
If every stage of a sequential quantum invocation records its outcome before the next stage is prepared, D(q₂⊛_Q q₁) = D(q₂)⊛D(q₁) — PRR-2026-003's Kleisli composition, recovered exactly.
Proof sketch.Recording every stage's outcome is exactly §4's decoherent case, applied at each stage in turn; no coherence survives between stages beyond what ⊛'s own law-of-total-probability composition already accounts for. ☐
Left open, matching PRR-2026-003 §7's own precedent exactly: whether the converse holds — that every quantum construction distinguishable from C under ⊛_Q requires an undecohered stage somewhere. Sufficiency is proven above; necessity is not, and is not claimed.
7. Invertibility Revisited
PRR-2026-003 Corollary 6.6 — the Unifying Theorem — proved that R, C, and the abstract nondeterministic relations A share exactly one group of invertible elements, G, embedded three different ways; the tour page states its conclusion plainly: "invertibility never gets richer just because things became probabilistic." This section reports where that claim stops holding.
Scope. Restricted here to the pre-measurement unitary sub-case of Q — a bare unitary preparation U_x : H_x → H_x at each x, with no measurement {Π_y^x} attached yet, composing under ordinary operator multiplication. Invertibility of the full measurement- inclusive triple is not analyzed here and is left as an explicit scope gap, not assumed away.
The invertible elements of this sub-case form U(H_x), the unitary group. For dim H_x ≥ 2, U(H_x) is a connected Lie group — a continuous one-parameter family of unitaries connects the identity to any other unitary, all invertible along the way. G = Sym(S) contains no nontrivial connected subgroup at all: it is built from bijections of a set, and no continuous path of bijections connects two distinct ones, however large G's own cardinality is. Hence Corollary 6.6's claim of one shared invertible group does not extend to this sub-case of Q — invertibility becomes structurally, not merely cardinality-wise, richer once amplitudes are allowed to vary.
This is stated as a correction against the series' own prior headline result, in the same spirit as PRR-2026-004's "which definition of deductive is more true" and PRR-2026-008's orthogonal-versus-contradiction correction — not smoothed over, not treated as though the original claim was always understood to be this narrowly scoped.
8. Geometry Retrospective
PRR-2026-008 Definition 6.1 gave reasoning operators an inner product: ⟨ρ₁,ρ₂⟩_{c,μ} = Σ_x μ(x)·Δ_c(ρ₁)(x)·Δ_c(ρ₂)(x), a real-valued, μ-weighted sum over S.
The map ρ ↦ (Δ_c(ρ)(x))_{x∈S} ∈ ℓ²(S,μ) identifies PRR-2026-008's inner product exactly with the real, phase-free, diagonal-basis special case of this paper's ⟨ψ₁|ψ₂⟩ structure — basis fixed to S itself, weights μ, every "amplitude" a real number (Δ_c(ρ)(x) itself, with no phase to speak of).
PRR-2026-008's entire geometry chapter needed no quantum apparatus to state. It is, in retrospect, precisely the shadow such apparatus casts once every amplitude is flattened to a real number and the basis fixed to S itself. Not claimed: that cos_{c,μ} refines Theorem 5.2's phase-dependent result, or says anything about which θ produces which score. Left open, not resolved by analogy.
9. Discussion and Conclusion
Most of this series survives the move to Q intact, recovered exactly the moment a quantum process is fully decohered: §4's Classical Reducibility, §6's consistency of the correctness functional, and Proposition 6.4's reduction of sequential composition. What does not survive is stated precisely, at the exact point the earlier proof stops reaching: §5's coherent pooling has no analogue of PRR-2026-008 Theorem 8.1, because its combinator never realizes two samples before combining them, and §7's invertibility result shows PRR-2026-003's Unifying Theorem was scoped to non-determinism over outcomes, not over amplitudes.
The AI Concretization
Following this series' established practice (PRR-2026-002 §11) of proposing a concrete model for what plays the role of these operators in an actual computational system: we claim that an AI reasoning system, weighing several candidate hypotheses before committing to a single output, is a Q-structured process — the candidate hypotheses form a coherent superposition, and the eventual committed output is the result of a Born-rule measurement against them. Under this reading, §5's interference result explains why two individually uninformative lines of reasoning can combine into either confident correctness or confident error depending on how they are weighed together, and §7's symmetry break explains why the space of ways to revise an uncommitted judgment is continuous where the space of ways to revise a committed answer is not.
Left open: the necessity direction of Proposition 6.4 (§6, matching PRR-2026-003 §7's own unresolved converse exactly), and whether cos_{c,μ} says anything about Theorem 5.2's phase dependence (§8). Neither is assumed resolved by this paper.
References
Edwards, R. (2026). Edwardian Algebra. Perspectivity Research Reports, Vol. 1, No. 2, PRR-2026-002.
Edwards, R. (2026). Concrete Inferential Algebra. Perspectivity Research Reports, Vol. 1, No. 3, PRR-2026-003.
Edwards, R. (2026). Correctness Gradient Algebra. Perspectivity Research Reports, Vol. 1, No. 5, PRR-2026-005.
Edwards, R. (2026). Correctness Geometry Algebra. Perspectivity Research Reports, Vol. 1, No. 8, PRR-2026-008.
Birkhoff, G., & von Neumann, J. (1936). The Logic of Quantum Mechanics. Annals of Mathematics, 37(4), 823–843.
Nielsen, M. A., & Chuang, I. L. (2010). Quantum Computation and Quantum Information (10th Anniversary Edition). Cambridge University Press.
Peer Review
Section 5 earns this paper's place in the series the same way PRR-2026-008's §8 did: it does not merely define a new object and admire it, it pushes the new object against an existing unconditional theorem and states precisely why that theorem's proof does not reach the new case. The decision to frame Theorem 5.2 as "not a counterexample to Theorem 8.1" rather than as a refutation is the correct one, and stated clearly. Recommend acceptance.
Section 7's willingness to state a correction against the series' own quoted headline result — rather than quietly scoping around it — is exactly the discipline this series has maintained since PRR-2026-004. The explicit scope boundary in §4 (commuting projectors proven classical; the harder Boolean-lattice claim cited but not reproved) is the right level of honesty about what was and wasn't established here. Recommend acceptance.