Algebra-Relative Branch Multiplicity in Decoherent Histories: Orbit Reduction, Record Compression, and Logical Recovery

1. Introduction

Everettian branch structure is useful only after one has specified the physical distinctions by which branches are individuated. A wavefunction may be expanded in many bases. Even a decoherence-selected expansion depends on environmental monitoring, stable records, and coarse-grained variables. Branch number is therefore not a primitive invariant of the universal quantum state.

This paper asks what is eligible to count as a distinct branch. The answer developed below has three layers. First, exact label redundancy is removed by physical symmetries. Second, histories are compared through an explicitly chosen finite-dimensional operator system of admissible effects. Third, in protected systems, histories that induce the same recovered logical channel are identified at the logical level. The claim is not literal world-merger. Unitary evolution does not map orthogonal global branches to a single pure state. The claim is algebraic: microscopic distinction need not imply effective branch distinction.

The central object is an effective branch packing number

$$N_{\mathrm{eff}}(t;\mathcal{A},\epsilon ),$$

(1)

the largest number of mutually $\epsilon$-distinguishable symmetry classes of histories visible to $\mathcal{A}$. Packing is used instead of approximate quotienting because approximate indistinguishability need not be transitive.

The main formal consequences are organized as a theorem cascade. Orbit-packing reduction shows that invariant records force quotienting by exact symmetry. Branch packing is monotone under refinement of $\mathcal{A}$. Recovery and other completely positive trace-preserving post-processing maps contract operational channel distinguishability. Collective-record models can reduce exponentially many labeled histories to polynomially many exchange classes and then to still smaller record classes. For N identical bosonic d-level systems monitored only through an equally spaced collective charge,

$$d^N⟶\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right)⟶(d-1)N+1,$$

(2)

where the arrows denote exact exchange quotienting and restriction to the collective record algebra. More generally, if the recorded collective variables span an r-dimensional linear record space, the record count grows at most as $O\left(N^r\right)$, while the fine labeled count grows as $d^N$.

Quantum error correction supplies the sharpest logical example. A syndrome measurement may record which physical error occurred while recovery leaves the logical channel unchanged. A syndrome distinction is then a record distinction, not a logical branch distinction. The paper tracks this through four counts,

$$N_{\mathrm{fine}},N_{\mathrm{sym}},N_{\mathrm{record}},N_{\mathrm{logical}},$$

(3)

for fine labeled histories, symmetry classes, record-visible histories, and recovered logical histories. Table 1 summarizes the branch-individuating question at each level.

Section 2 positions the proposal relative to decoherent histories, Diósi’s finite-dimensional bound, Saunders’s branch-counting rule, Quantum Darwinism, and quantum error correction. Section 3 defines symmetry quotienting of candidate histories. Section 4 defines algebra-relative branch packing and proves monotonicity, a packing bound, and a collective-record compression theorem. Section 5 proves recovery contraction and the recoverability-collapse criterion. Section 6 gives examples: the three-qubit bit-flip code, identical bosonic qudit occupation sectors, and an exchange-quotiented history-space model for dipositronium annihilation records. Section 7 briefly discusses partial recoherence and time-dependent effective multiplicity.

2. Prior Work and Positioning

The paper does not claim that branch structure is newly shown to be emergent, approximate, or record-dependent. That lesson is already central to decoherent histories, decoherence theory, Everettian interpretations, and Quantum Darwinism.

In decoherent or consistent histories, histories are represented by time-ordered chains of projectors and assigned probabilities when interference between alternatives is sufficiently suppressed [2,3,4,5]. The histories that enter the formalism are already coarse-grained. Gell-Mann and Hartle emphasize that suitable recorded histories can support ordinary probabilistic description, and later work refines the relation between records, decoherence, and quasi-classical domains [11,12,13].

Diósi gives the nearest formal precedent for bounded multiplicity. For an N-state quantum system, he obtained an $N^2$ upper bound on the number of histories in a decohering family [6]. The present paper addresses a different issue: which candidate histories should be treated as distinct before such a count is attempted.

Saunders gives the closest overlap on Everettian branch counting. His branch-counting rule is state-dependent, uses decoherent histories, and explicitly connects branch counting with the combinatorics of Planck, Bose, Einstein, and Dirac [14]. The distinction here is narrower: Saunders asks how branches may be counted once a branching structure is supplied; the present paper asks what must first be identified before branch-counting begins.

Quantum Darwinism supplies the record criterion. Blume-Kohout and Zurek argue that classical objectivity is tied to redundantly stored environmental information [8]. Riedel, Zurek, and Zwolak show that such redundancy can rise and later be degraded by environmental interactions [9]. These results support a record-relative and time-dependent notion of effective branch multiplicity.

Quantum error correction supplies the logical criterion. The Knill-Laflamme conditions characterize correctable error sets and recovery operations on an encoded subspace [15,16,17]. The present use of error correction is operational: histories that induce the same recovered logical channel are not distinct logical branches at the resolution of the code algebra.

Identical-particle quantum mechanics motivates the exact symmetry quotient. Permutation-related labeled descriptions of identical constituents are not automatically distinct physical alternatives. In occupation-number language, the physical alternative is the exchange-invariant sector or occupation pattern, not the arbitrary particle label.

3. Candidate Histories and Exact Symmetry Quotienting

Let $\mathcal{B}$ be a finite set of candidate histories drawn from a decoherent or approximately decoherent family. The construction below refines the individuation of histories that have already passed the usual decoherence or record-admissibility requirements; it is not intended to pack arbitrary incompatible chains of projections.

A history $h\in \mathcal{B}$ is represented by a class operator

$$C_h=P_{{\alpha }_n}\left(t_n\right)\dots P_{{\alpha }_1}\left(t_1\right),$$

(4)

where the $P_{{\alpha }_k}\left(t_k\right)$ are Heisenberg-picture projectors or coarse-grained alternatives at times $t_k$. For an initial density operator ${\rho }_0$, set

$${\tilde{\rho }}_h=C_h{\rho }_0{C}_h^{†},p_h=Tr{\tilde{\rho }}_h.$$

(5)

When $p_h>0$, the normalized branch state is

$${\rho }_h={\displaystyle \frac{{\tilde{\rho }}_h}{p_h}}.$$

(6)

The naive fine count is

$$N_{\mathrm{fine}}=\left|\mathcal{B}\right|.$$

(7)

This number can overcount physical alternatives if $\mathcal{B}$ is written in a labeled constituent space.

Let G be an exact physical symmetry group represented by unitaries $U_g$. It acts on class operators by

$$C_h\mapsto U_g{C}_h{U}_g^{†}.$$

(8)

For identical particles, G may be a permutation group such as $S_N$. For dipositronium, the relevant exchange symmetry includes $S_2^{\left(e^{-}\right)}\times S_2^{\left(e^{+}\right)}$.

Let

$${\mathcal{A}}_G=\{O:[O,U_g]=0\mathrm{for}\mathrm{all}g\in G\}$$

(9)

be the invariant observable algebra. Define $h{\sim }_G{h}^{\prime }$ when

$$C_{h^{\prime }}=U_g{C}_h{U}_g^{†}$$

(10)

for some $g\in G$, relative to the invariant algebra ${\mathcal{A}}_G$. The symmetry-quotiented history set is

$${\mathcal{B}}_G=\mathcal{B}/G,$$

(11)

and the symmetry-reduced count is

$$N_{\mathrm{sym}}=\left|{\mathcal{B}}_G\right|.$$

(12)

Lemma 1 (Exact symmetry quotient).

Let $h,h^{\prime }\in \mathcal{B}$ be related by a physical symmetry $g\in G$:

$$C_{h^{\prime }}=U_g{C}_h{U}_g^{†}.$$

(13)

Assume $U_g{\rho }_0{U}_g^{†}={\rho }_0$ and let $O\in {\mathcal{A}}_G$. Then

$$Tr\left(O{\tilde{\rho }}_h\right)=Tr\left(O{\tilde{\rho }}_{h^{\prime }}\right).$$

(14)

If $p_h,p_{h^{\prime }}>0$, then

$$Tr\left(O{\rho }_h\right)=Tr\left(O{\rho }_{h^{\prime }}\right).$$

(15)

Thus h and $h^{\prime }$ are not distinct physical branches relative to ${\mathcal{A}}_G$.

Proof. 

The assumptions give

$$\begin{array}{cc}\hfill {\tilde{\rho }}_{h^{\prime }}& =C_{h^{\prime }}{\rho }_0{C}_{h^{\prime }}^{†}\hfill \\ \hfill & =U_g{C}_h{U}_g^{†}{\rho }_0{U}_g{C}_h^{†}{U}_g^{†}\hfill \\ \hfill & =U_g{C}_h{\rho }_0{C}_h^{†}{U}_g^{†}=U_g{\tilde{\rho }}_h{U}_g^{†}.\hfill \end{array}$$

(16)

For $O\in {\mathcal{A}}_G$,

$$\begin{array}{cc}\hfill Tr\left(O{\tilde{\rho }}_{h^{\prime }}\right)& =Tr\left(O{U}_g{\tilde{\rho }}_h{U}_g^{†}\right)\hfill \\ \hfill & =Tr\left(U_g^{†}O{U}_g{\tilde{\rho }}_h\right)\hfill \\ \hfill & =Tr\left(O{\tilde{\rho }}_h\right).\hfill \end{array}$$

(17)

Taking $O=I$ gives $p_h=p_{h^{\prime }}$, and the normalized statement follows. □

Lemma 1 is elementary, but it prevents a common error: labeled alternatives related by an exact physical symmetry should not be counted as distinct branches for invariant observers or records.

4. Algebra-Relative Branch Packing

Exact symmetry quotienting removes label redundancy. The next issue is distinguishability relative to the measurements available at a given level of description. Throughout, $\mathcal{A}$ denotes a finite-dimensional unital operator system of admissible effects. When these effects correspond to stable detector, environmental, syndrome, or logical records, $\mathcal{A}$ will be called a record algebra. In special cases $\mathcal{A}$ may be a $C^{*}$-algebra, but the definitions below require only the ordered effect set $0\le M\le I$ in $\mathcal{A}$.

Examples include a pointer operator system, an occupation-number algebra, an environmental-record algebra, a photon-channel record algebra, a syndrome algebra, or a recovered logical algebra. When $\mathcal{A}$ is used on the quotient ${\mathcal{B}}_G$, its effects are assumed to descend to that quotient, for example by being G-invariant. If a physically admissible record system genuinely distinguishes the labels, those labels are not redundant relative to that record system.

Define distinguishability directly by the effects in $\mathcal{A}$. For histories $h,h^{\prime }\in {\mathcal{B}}_G$, let

$${\Delta }_{\mathcal{A}}(h,h^{\prime };t)=\underset{0\le M\le I,M\in \mathcal{A}}{sup}\left|Tr\left[M\left({\rho }_h\left(t\right)-{\rho }_{h^{\prime }}\left(t\right)\right)\right]\right|.$$

(18)

This is the maximum measurement bias available to effects in $\mathcal{A}$.

Theorem 1 (Orbit-packing reduction).

Let G act on $\mathcal{B}$, and let $\mathcal{A}$ be a G-invariant admissible operator system. If h and $h^{\prime }$ lie in the same G-orbit, then

$${\Delta }_{\mathcal{A}}(h,h^{\prime };t)=0.$$

(19)

Consequently, no ε-packing for $\epsilon >0$ contains two members of the same orbit. Equivalently, the effective branch packing number computed on the labeled history set equals the packing number computed on the quotient set $\mathcal{B}/G$ with the induced pseudometric:

$$N_{\mathrm{eff}}^{\mathcal{B}}(t;\mathcal{A},\epsilon )=N_{\mathrm{eff}}^{\mathcal{B}/G}(t;\mathcal{A},\epsilon ).$$

(20)

Proof. 

By Lemma 1, all effects in the invariant operator system have identical expectation values on normalized branch states in the same orbit. The supremum in Eq. (18) is therefore zero. Hence an $\epsilon$-separated set can contain at most one representative from each orbit.

Every $\epsilon$-separated subset of the labeled set maps injectively to an $\epsilon$-separated subset of the quotient, since no two elements of one orbit can both appear. Conversely, any $\epsilon$-separated subset of the quotient has a choice of representatives in the labeled set, and those representatives are $\epsilon$-separated because the quotient pseudometric is induced from ${\Delta }_{\mathcal{A}}$. The maximal cardinalities coincide. □

If $\mathcal{A}=\mathcal{B}\left(\mathcal{H}\right)$, then

$${\Delta }_{\mathcal{B}\left(\mathcal{H}\right)}(h,h^{\prime };t)={\displaystyle \frac{1}{2}}{\parallel {\rho }_h\left(t\right)-{\rho }_{h^{\prime }}\left(t\right)\parallel }_1.$$

(21)

For a proper subalgebra or operator system, Eq. (18) is the corresponding restricted operational distinguishability.

Approximate indistinguishability is not generally transitive. For that reason, define effective multiplicity by mutual separation rather than by an approximate quotient.

Definition 1 (Effective branch packing number).

For resolution $\epsilon >0$,

$$N_{\mathrm{eff}}(t;\mathcal{A},\epsilon )=max\left\{\left|S\right|:S\subseteq {\mathcal{B}}_G,{\Delta }_{\mathcal{A}}(h,h^{\prime };t)>\epsilon \mathrm{for}\mathrm{all}h\ne h^{\prime }\in S\right\}.$$

(22)

Thus $N_{\mathrm{eff}}(t;\mathcal{A},\epsilon )$ is the largest number of mutually distinguishable symmetry classes of histories visible to $\mathcal{A}$ at resolution $\epsilon$. For infinite families one may replace the maximum by a supremum; the finite case is enough for the examples below.

Proposition 1 (Monotonicity under algebra refinement).

Let ${\mathcal{A}}_0\subseteq {\mathcal{A}}_1$ be two admissible operator systems acting on the same branch states. Then

$${\Delta }_{{\mathcal{A}}_0}(h,h^{\prime };t)\le {\Delta }_{{\mathcal{A}}_1}(h,h^{\prime };t)$$

(23)

for all histories $h,h^{\prime }$. Consequently,

$$N_{\mathrm{eff}}(t;{\mathcal{A}}_0,\epsilon )\le N_{\mathrm{eff}}(t;{\mathcal{A}}_1,\epsilon ).$$

(24)

Proof. 

The effects in ${\mathcal{A}}_0$ form a subset of the effects in ${\mathcal{A}}_1$. The supremum in Eq. (18) can therefore only increase when ${\mathcal{A}}_0$ is replaced by ${\mathcal{A}}_1$. If $S\subseteq {\mathcal{B}}_G$ is $\epsilon$-separated with respect to ${\Delta }_{{\mathcal{A}}_0}$, then it is also $\epsilon$-separated with respect to ${\Delta }_{{\mathcal{A}}_1}$. Taking the maximum over such S proves the claim. □

Corollary 1 (Algebraic hierarchy of branch counts).

For a chain of operator systems

$${\mathcal{A}}_{\mathrm{logical}}\subseteq {\mathcal{A}}_{\mathrm{record}}\subseteq {\mathcal{A}}_{\mathrm{fine}},$$

(25)

one has

$$N_{\mathrm{eff}}(t;{\mathcal{A}}_{\mathrm{logical}},\epsilon )\le N_{\mathrm{eff}}(t;{\mathcal{A}}_{\mathrm{record}},\epsilon )\le N_{\mathrm{eff}}(t;{\mathcal{A}}_{\mathrm{fine}},\epsilon ).$$

(26)

This corollary gives a mathematical version of the four-level hierarchy. Finer algebras can distinguish at least as many branch classes as coarser algebras. A syndrome algebra can therefore separate histories that a recovered logical algebra identifies.

Proposition 2 (Algebraic packing bound).

Let ${\mathcal{S}}_{\mathcal{A}}$ be the compact state space obtained from measurement statistics of effects in $\mathcal{A}$. Then $N_{\mathrm{eff}}(t;\mathcal{A},\epsilon )$ is bounded above by the ε-packing number of ${\mathcal{S}}_{\mathcal{A}}$ in the metric induced by Eq. (18). If $\mathcal{A}$ is finite-dimensional and $d_{\mathcal{A}}$ is the real affine dimension of ${\mathcal{S}}_{\mathcal{A}}$, then for a norm-dependent constant $C_{\mathcal{A}}$,

$$N_{\mathrm{eff}}(t;\mathcal{A},\epsilon )\le {\left({\displaystyle \frac{C_{\mathcal{A}}}{\epsilon }}\right)}^{d_{\mathcal{A}}}.$$

(27)

Proof 

(Proof sketch). The branch states restricted to the measurement statistics of $\mathcal{A}$ form a finite subset of the compact convex set ${\mathcal{S}}_{\mathcal{A}}$. An $\epsilon$-separated subset of those restricted states is an $\epsilon$-packing. Standard finite-dimensional covering or volume estimates give a bound of the displayed form, with the constant depending on the chosen norm and normalization. □

Proposition 2 is a general metric statement. The next result gives a concrete branch-count compression for collective qudit records.

Theorem 2 (Collective linear-record compression).

Consider N identical bosonic d-level constituents with occupation vectors

$${\mathsf{\Omega }}_{N,d}=\left\{(n_0,\dots ,n_{d-1}):n_{\alpha }\in \mathbb{N},{\displaystyle \sum _{\alpha =0}^{d-1}}{n}_{\alpha }=N\right\}.$$

(28)

Let m commuting collective record observables be

$$S^{\left(r\right)}={\displaystyle \sum _{\alpha =0}^{d-1}}{g}_{\alpha }^{\left(r\right)}{N}_{\alpha },r=1,\dots ,m,$$

(29)

with integer weights $g_{\alpha }^{\left(r\right)}$. Let $G_{r\alpha }=g_{\alpha }^{\left(r\right)}$ be the resulting $m\times d$ record matrix and let $r_0={rank}_{\mathbb{R}}(G-G_0)$ be the affine rank of the column differences, where $G_0$ denotes any fixed column of G repeated d times. If the record operator system resolves only the joint values of the $S^{\left(r\right)}$, then

$$N_{\mathrm{record}}=\left|G{\mathsf{\Omega }}_{N,d}\right|\le {\displaystyle \prod _{r=1}^m}(D_{r}N+1),D_r=\underset{\alpha }{max}{g}_{\alpha }^{\left(r\right)}-\underset{\alpha }{min}{g}_{\alpha }^{\left(r\right)}.$$

(30)

Moreover, for fixed $d,m,G$,

$$N_{\mathrm{record}}=O\left(N^{r_0}\right).$$

(31)

If $r_0>0$, this exponent is sharp: there exists a constant $c>0$, depending only on G and d, such that for all sufficiently large N,

$$N_{\mathrm{record}}\ge c{N}^{r_0}.$$

(32)

Consequently,

$$N_{\mathrm{record}}=\mathsf{\Theta }\left(N^{r_0}\right)$$

(33)

for nonzero affine rank. In contrast, the fine labeled count is $d^N$ and the exact bosonic exchange count is

$$N_{\mathrm{sym}}=\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right).$$

(34)

Proof. 

The exchange-quotiented alternatives are the occupation vectors in ${\mathsf{\Omega }}_{N,d}$, whose cardinality is the stars-and-bars value $\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right)$. For each record component r, the value of $S^{\left(r\right)}$ lies between

$$N\underset{\alpha }{min}{g}_{\alpha }^{\left(r\right)}\mathrm{and}N\underset{\alpha }{max}{g}_{\alpha }^{\left(r\right)}.$$

(35)

Since the weights are integral, the number of possible values of this component is at most $D_{r}N+1$. The joint value lies in the Cartesian product of these finite ranges, giving Eq. (30).

For the upper rank statement, $G{\mathsf{\Omega }}_{N,d}$ is contained in the set of integer points of the dilation by N of the polytope $conv\{G{e}_0,\dots ,G{e}_{d-1}\}$. This polytope has affine dimension $r_0$. A fixed $r_0$-dimensional polytope has $O\left(N^{r_0}\right)$ lattice points in its N-fold dilation. Hence $|G{\mathsf{\Omega }}_{N,d}|=O\left(N^{r_0}\right)$.

For the lower bound, choose $r_0+1$ columns $v_0,v_1,\dots ,v_{r_0}$ of G that are affinely independent. For every integer vector $(k_1,\dots ,k_{r_0})$ with $k_i\ge 0$ and ${\sum }_i{k}_i\le N$, the occupation vector that places $k_i$ particles in level $v_i$, $N-{\sum }_i{k}_i$ particles in level $v_0$, and none in the other selected levels gives the record value

$$N{v}_0+{\displaystyle \sum _{i=1}^{r_0}}{k}_i(v_i-v_0).$$

(36)

Affine independence makes these values distinct. The number of such integer $r_0$-tuples is $\left({\displaystyle \genfrac{}{}{0pt}{}{N+r_0}{r_0}}\right)=\mathsf{\Theta }\left(N^{r_0}\right)$. This proves the lower bound and hence the sharp asymptotic growth when $r_0>0$. □

Corollary 2 (Equally spaced one-charge record).

For a single collective charge

$$S={\displaystyle \sum _{\alpha =0}^{d-1}}\alpha N_{\alpha },$$

(37)

the record-visible count is exactly

$$N_{\mathrm{record}}=(d-1)N+1.$$

(38)

Thus

$$d^N⟶\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right)⟶(d-1)N+1.$$

(39)

Proof. 

The upper bound follows from Theorem 2 with $m=1$ and $D_1=d-1$. Every integer k with $0\le k\le (d-1)N$ can be written as a sum of N integers from $\{0,1,\dots ,d-1\}$: take as many $(d-1)$’s as possible, one remainder if needed, and zeros for the rest. Hence every value in the interval occurs. □

When the detector has finite resolution, exact joint eigenspaces are replaced by bins or noisy record distributions. Then $N_{\mathrm{record}}$ is replaced by the corresponding $\epsilon$-packing number of the record distributions. This makes the packing definition operational rather than merely nominal.

5. Recoverability-Limited Branch Distinction

Quantum error correction makes precise a distinction that decoherence alone does not: a history may be recorded without being logically distinct after recovery. This section states that distinction at the channel level.

5.1. Code Algebra and History Maps

Let ${\mathcal{H}}_C$ be a finite-dimensional code space with projector P embedded in a physical Hilbert space ${\mathcal{H}}_{\mathrm{phys}}$. In finite dimension we identify logical density operators with elements of

$$\mathcal{C}=\mathcal{B}\left({\mathcal{H}}_C\right),$$

(40)

or equivalently with operators supported on $P{\mathcal{H}}_{\mathrm{phys}}$.

A microscopic history h induces a completely positive branch map

$${\mathcal{N}}_h:\mathcal{B}\left({\mathcal{H}}_C\right)⟶\mathcal{B}\left({\mathcal{H}}_{\mathrm{phys}}\right).$$

(41)

In deterministic examples this map is trace preserving. For conditioned branch histories it may be trace non-increasing before normalization; the channel-level contraction result below is stated for trace-preserving maps, while the correctable branch example uses normalized successful branches.

Let

$$\mathcal{R}:\mathcal{B}\left({\mathcal{H}}_{\mathrm{phys}}\right)⟶\mathcal{B}\left({\mathcal{H}}_C\right)$$

(42)

be a recovery channel. The recovered logical map associated with h is

$${\mathsf{\Phi }}_h=(\mathcal{R}\circ {\mathcal{N}}_h){|}_{\mathcal{C}}.$$

(43)

The notation ${|}_{\mathcal{C}}$ means that the comparison is restricted to encoded input states, or equivalently to the induced action on the logical operator space. The operational norm used below is the diamond norm on this restricted input domain.

5.2. Contraction Under Recovery

Proposition 3 (Recovery contraction).

Let ${\mathcal{N}}_h$ and ${\mathcal{N}}_{h^{\prime }}$ be two channels on the code-input domain, and let $\mathcal{R}$ be a completely positive trace-preserving recovery channel. Then

$$\parallel \mathcal{R}\circ {\mathcal{N}}_h-\mathcal{R}\circ {\mathcal{N}}_{h^{\prime }}{\parallel }_{\diamond }\le {\parallel {\mathcal{N}}_h-{\mathcal{N}}_{h^{\prime }}\parallel }_{\diamond }.$$

(44)

Consequently, post-processing by recovery cannot increase channel distinguishability.

Proof. 

The diamond norm is submultiplicative under composition:

$$\parallel \mathcal{R}\circ ({\mathcal{N}}_h-{\mathcal{N}}_{h^{\prime }}){\parallel }_{\diamond }\le {\parallel \mathcal{R}\parallel }_{\diamond }{\parallel {\mathcal{N}}_h-{\mathcal{N}}_{h^{\prime }}\parallel }_{\diamond }.$$

(45)

For a completely positive trace-preserving map, ${\parallel \mathcal{R}\parallel }_{\diamond }=1$. This gives the result. □

This is a data-processing form of branch compression. Recovery does not merely repair states; it can reduce the operational distinguishability of microscopic history channels when the comparison is made at the logical output.

5.3. Recoverability-Collapse Theorem

Theorem 3 (Recoverability-collapse theorem).

Let $\mathcal{F}\subseteq \mathcal{B}$ be a family of microscopic histories inducing recovered logical maps

$${\mathsf{\Phi }}_h=(\mathcal{R}\circ {\mathcal{N}}_h){|}_{\mathcal{C}},h\in \mathcal{F}.$$

(46)

Suppose there exists a logical channel $\mathsf{\Lambda }:\mathcal{C}\to \mathcal{C}$ such that

$$\parallel {\mathsf{\Phi }}_h{-\mathsf{\Lambda }\parallel }_{\diamond }\le \eta \mathrm{for}\mathrm{all}h\in \mathcal{F}.$$

(47)

Then, for all $h,h^{\prime }\in \mathcal{F}$,

$$\parallel {\mathsf{\Phi }}_h-{\mathsf{\Phi }}_{h^{\prime }}{\parallel }_{\diamond }\le 2\eta .$$

(48)

Consequently, if $2\eta \le \epsilon$, the family $\mathcal{F}$ contributes at most one recovered logical branch at resolution ε.

Proof. 

For any $h,h^{\prime }\in \mathcal{F}$, the triangle inequality gives

$$\begin{array}{cc}\hfill \parallel {\mathsf{\Phi }}_h-{\mathsf{\Phi }}_{h^{\prime }}{\parallel }_{\diamond }& =\parallel {\mathsf{\Phi }}_h-\mathsf{\Lambda }+\mathsf{\Lambda }-{\mathsf{\Phi }}_{h^{\prime }}{\parallel }_{\diamond }\hfill \\ \hfill & \le \parallel {\mathsf{\Phi }}_h{-\mathsf{\Lambda }\parallel }_{\diamond }+{\parallel \mathsf{\Lambda }-{\mathsf{\Phi }}_{h^{\prime }}\parallel }_{\diamond }\hfill \\ \hfill & \le 2\eta .\hfill \end{array}$$

(49)

If $2\eta \le \epsilon$, no two members of $\mathcal{F}$ are $\epsilon$-separated in the recovered logical channel metric. Hence an $\epsilon$-packing of recovered logical branches contains at most one representative from $\mathcal{F}$. □

Theorem 4 (Logical cluster packing theorem).

Let a finite family of microscopic histories decompose into nonempty clusters

$$\mathcal{F}=C_1\cup \dots \cup C_m.$$

(50)

Suppose that for each cluster $C_j$ there exists a logical channel ${\mathsf{\Lambda }}_j$ such that

$$\parallel {\mathsf{\Phi }}_h-{\mathsf{\Lambda }}_j{\parallel }_{\diamond }\le \eta \mathrm{for}\mathrm{all}h\in C_j.$$

(51)

Assume also that $2\eta \le \epsilon$ and that the cluster centers are separated by

$$\parallel {\mathsf{\Lambda }}_i-{\mathsf{\Lambda }}_j{\parallel }_{\diamond }>\epsilon +2\eta (i\ne j).$$

(52)

Then the recovered logical branch packing number of $\mathcal{F}$ at resolution ε is exactly m.

Proof. 

By Theorem 3, no $\epsilon$-packing can contain more than one history from any cluster $C_j$. Hence the packing number is at most m.

Choose one representative $h_j\in C_j$ for each j. For $i\ne j$, the reverse triangle inequality gives

$$\begin{array}{cc}\hfill \parallel {\mathsf{\Phi }}_{h_i}-{\mathsf{\Phi }}_{h_j}{\parallel }_{\diamond }& \ge \parallel {\mathsf{\Lambda }}_i-{\mathsf{\Lambda }}_j{\parallel }_{\diamond }-\parallel {\mathsf{\Phi }}_{h_i}-{\mathsf{\Lambda }}_i{\parallel }_{\diamond }-{\parallel {\mathsf{\Phi }}_{h_j}-{\mathsf{\Lambda }}_j\parallel }_{\diamond }\hfill \\ \hfill & >(\epsilon +2\eta )-\eta -\eta =\epsilon .\hfill \end{array}$$

(53)

Thus the chosen representatives are pairwise $\epsilon$-separated, so the packing number is at least m. The two bounds coincide. □

This cluster theorem is the channel-level version of the intended branch criterion: recovered logical branch multiplicity is the number of distinguishable recovered logical-channel clusters, not the number of microscopic histories in those clusters.

The use made of Theorem 3 is the branch-individuation criterion: histories that induce indistinguishable recovered logical maps should not be counted as distinct logical branches. In particular,

$$\mathrm{syndrome}-\mathrm{distinct}\nRightarrow \mathrm{logical}-\mathrm{branch}-\mathrm{distinct}$$

(54)

is not a valid inference. A syndrome distinction is a record distinction; it need not be a logical distinction.

5.4. Knill-Laflamme Corollary

Corollary 3 (Correctable errors induce one recovered logical branch).

Let P be a code projector correcting an error set $\left\{E_a\right\}$ satisfying the Knill-Laflamme condition

$$P{E}_a^{†}{E}_{b}P={\lambda }_{ab}P.$$

(55)

Equivalently, after a unitary change of Kraus basis, one may use a correctable error family $\left\{F_{\mu }\right\}$ for which

$$P{F}_{\mu }^{†}{F}_{\nu }P=d_{\mu }{\delta }_{\mu \nu }P.$$

(56)

For the successful correctable branch family associated with these syndrome-resolved errors, all branches induce the same recovered logical channel on the code, namely the identity channel in the ideal case.

Proof sketch.

Condition (55) is necessary and sufficient for correctability of the error set on the code. Diagonalizing the positive matrix ${\lambda }_{ab}$ gives an equivalent Kraus family $\left\{F_{\mu }\right\}$ with orthogonal error subspaces, up to null errors. There is then a syndrome measurement and syndrome-conditioned recovery such that, for encoded states $\rho =P\rho P$ and a syndrome-resolved correctable branch $\mu$,

$${\mathcal{R}}_{\mu }\left(F_{\mu }\rho F_{\mu }^{†}\right)=d_{\mu }\rho ,$$

(57)

or, after conditioning on that branch and normalizing, $\rho$. Thus every successful correctable branch has recovered logical map ${id}_{\mathcal{C}}$. Theorem 3 applies with $\eta =0$ and $\mathsf{\Lambda }={id}_{\mathcal{C}}$. □

The corollary is conditional on the successful correctable family and on a syndrome-resolved branch description. Uncorrectable multi-error histories, failed correction outcomes, or records that are not part of the chosen branch family are outside this toy count unless added explicitly and compared with the same algebra-relative method.

6. Worked Examples

The examples are not independent derivations of the Everett interpretation. They show how the formal hierarchy changes branch counts in concrete settings.

6.1. Three-Qubit Bit-Flip Code: Record Branches Versus Logical Branches

The bit-flip code is the flagship recovery example. It is not an identical-particle exchange example; the three qubits are distinguishable physical subsystems. Its purpose is to separate syndrome records from recovered logical branches.

The code is

$$|0_L\rangle =|000\rangle ,|1_L\rangle =|111\rangle .$$

(58)

Restrict attention to the successful correctable family: no error and one bit-flip error. The corresponding branch channels on code states are

$${\mathcal{N}}_0\left(\rho \right)=\rho ,{\mathcal{N}}_i\left(\rho \right)=X_i\rho X_i,i=1,2,3.$$

(59)

The syndrome measurement has four outcomes, corresponding to

$$I,X_1,X_2,X_3.$$

(60)

Thus the syndrome algebra distinguishes four record branches:

$$N_{\mathrm{record}}=4.$$

(61)

The recovery is syndrome-conditioned:

$${\mathcal{R}}_0\left(\rho \right)=\rho ,{\mathcal{R}}_i\left(\rho \right)=X_i\rho X_i,i=1,2,3.$$

(62)

For any code state ${\rho }_L$,

$$({\mathcal{R}}_i\circ {\mathcal{N}}_i)\left({\rho }_L\right)=X_i\left(X_i{\rho }_L{X}_i\right)X_i={\rho }_L.$$

(63)

Hence

$${\mathsf{\Phi }}_i={id}_{\mathcal{C}}(i=0,1,2,3),$$

(64)

and the recovered logical algebra sees one successful logical branch:

$$N_{\mathrm{logical}}=1.$$

(65)

The same correctable family therefore has

$$N_{\mathrm{record}}=4,N_{\mathrm{logical}}=1.$$

(66)

The syndrome register remembers which error occurred; the recovered logical qubit does not.

6.2. Identical Bosonic Qudits and Collective-Record Scaling

Consider N identical bosonic d-level constituents with one-particle levels labelled $0,1,\dots ,d-1$. A labeled tensor-product description gives

$$N_{\mathrm{fine}}=d^N.$$

(67)

The physical bosonic occupation sectors are

$$(n_0,n_1,\dots ,n_{d-1}),{\displaystyle \sum _{\alpha =0}^{d-1}}{n}_{\alpha }=N,$$

(68)

so exact exchange quotienting gives

$$N_{\mathrm{sym}}=\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right).$$

(69)

For fixed d, this is polynomial in N, whereas the labeled count is exponential.

Now take the record algebra to be generated by the collective charge

$$S={\displaystyle \sum _{\alpha =0}^{d-1}}\alpha N_{\alpha }.$$

(70)

Corollary 2 gives

$$N_{\mathrm{record}}=(d-1)N+1.$$

(71)

The scaling is therefore

$$d^N⟶\left({\displaystyle \genfrac{}{}{0pt}{}{N+d-1}{d-1}}\right)⟶(d-1)N+1.$$

(72)

This example gives an explicit exponential-to-polynomial-to-linear reduction.

For $d=3$,

$$N_{\mathrm{fine}}=3^N,N_{\mathrm{sym}}=\left({\displaystyle \genfrac{}{}{0pt}{}{N+2}{2}}\right)={\displaystyle \frac{(N+1)(N+2)}{2}},N_{\mathrm{record}}=2N+1.$$

(73)

For $N=4$ this becomes

$$81⟶15⟶9.$$

(74)

The multi-record theorem gives a second payoff. If two independent collective records are monitored, for example total excitation and a second weighted charge, the record count is bounded by a quadratic polynomial in N rather than by $d^N$. More generally, if the joint record map has affine rank $r_0$, the record count is $O\left(N^{r_0}\right)$. The packing count therefore reflects the number and resolution of recorded collective variables, not the microscopic Hilbert-space tensor labels.

If the record of S is noisy, the exact value $(d-1)N+1$ is replaced by the corresponding $\epsilon$-packing number of the noisy record distributions. For a detector that bins the charge values into intervals of width w, the ideal sharp-record count is reduced to at most

$$⌈{\displaystyle \frac{(d-1)N+1}{w}}⌉.$$

(75)

This is a direct use of the packing definition: different occupation sectors count as different record branches only when their induced record distributions are separated at the chosen resolution.

6.3. Dipositronium: Exchange-Quotiented Annihilation Records

Dipositronium supplies a physical case in which labeled-channel bookkeeping can overcount candidate histories. The purpose of this subsection is limited. It does not derive molecular annihilation rates and does not make the dipositronium model carry the recovery theorem. It gives a history-space construction showing how artificial pair labels are removed by exchange symmetry and then further coarse-grained by detector resolution.

Consider a $P{s}_2$-like system with two electrons and two positrons. In an auxiliary labeled bookkeeping space write the electron labels as

$$e_1^{-},e_2^{-}$$

(76)

and the positron labels as

$$e_1^{+},e_2^{+}.$$

(77)

A labeled description then suggests four electron-positron pairings

$$(i,j)\in \{1,2\}\times \{1,2\}.$$

(78)

Let the photon channel be

$$c\in \{2\gamma ,3\gamma \}.$$

(79)

At the labeled bookkeeping level introduce candidate jump histories

$$h_{ij}^c,i,j\in \{1,2\},c\in \{2\gamma ,3\gamma \}.$$

(80)

The labeled jump-history set is

$${\mathcal{B}}_{\mathrm{lab}}^{\mathrm{jump}}=\{h_{ij}^c:i,j=1,2,c\in \{2\gamma ,3\gamma \}\},$$

(81)

so that

$$N_{\mathrm{fine}}^{\mathrm{jump}}=8.$$

(82)

If the no-jump history $h_{⌀}$ is included, then

$${\mathcal{B}}_{\mathrm{lab}}={\mathcal{B}}_{\mathrm{lab}}^{\mathrm{jump}}\cup \left\{h_{⌀}\right\},N_{\mathrm{fine}}=9.$$

(83)

The distinction between these two counts is useful: 8 counts jump-only labeled histories, while 9 counts the same jump histories together with the global no-jump history.

The relevant exchange group is

$$G=S_2^{\left(e^{-}\right)}\times S_2^{\left(e^{+}\right)},$$

(84)

acting by

$$(\sigma ,\tau ):h_{ij}^c\mapsto h_{\sigma \left(i\right)\tau \left(j\right)}^c.$$

(85)

For fixed photon channel c, this action is transitive on the four labeled pair histories. Hence

$$\{h_{11}^c,h_{12}^c,h_{21}^c,h_{22}^c\}$$

(86)

is one exchange orbit. The no-jump history is a singleton orbit. Thus the exchange quotient gives

$${\mathcal{B}}_G^{\mathrm{jump}}=\{\left[2\gamma \right],\left[3\gamma \right]\},N_{\mathrm{sym}}^{\mathrm{jump}}=2,$$

(87)

and, with no-jump included,

$${\mathcal{B}}_G=\{\left[h_{⌀}\right],\left[2\gamma \right],\left[3\gamma \right]\},N_{\mathrm{sym}}=3.$$

(88)

The first reduction is therefore

$$8⟶2$$

(89)

for jump-only histories, or

$$9⟶3$$

(90)

when the no-jump history is included. This is an exact symmetry reduction, not an approximation to the annihilation dynamics.

The same structure can be expressed at the level of class operations. Let $J_{ij}^c$ denote a labeled bookkeeping jump operator for annihilation of the labeled pair $(i,j)$ into photon channel c. Exchange covariance means that, for $g=(\sigma ,\tau )$,

$$U_g{J}_{ij}^c{U}_g^{†}=J_{\sigma \left(i\right)\tau \left(j\right)}^c.$$

(91)

A record system invariant under G cannot use the artificial pair label $(i,j)$ as a branch-individuating variable. If photon channel is recorded but pair label is not, the recorded outcomes are

$$r\in \{⌀,2\gamma ,3\gamma \}.$$

(92)

There are two useful ways to write the unresolved jump operation, depending on what is assumed about unobserved degrees of freedom. If the hidden environment distinguishes the labeled alternatives, the channel-c instrument element is the CP map

$${\mathcal{H}}_c\left(\rho \right)={\displaystyle \sum _{i,j=1}^2}{J}_{ij}^c\rho J_{ij}^{c†}.$$

(93)

If the exchange-summed coherent idealization is appropriate, one may instead define

$$J_c^{\mathrm{ex}}={\displaystyle \sum _{i,j=1}^2}{J}_{ij}^c$$

(94)

and write

$${\mathcal{H}}_c^{\mathrm{coh}}\left(\rho \right)=J_c^{\mathrm{ex}}\rho J_c^{\mathrm{ex}†}.$$

(95)

Equations (93) and (95) are not asserted to be dynamically equivalent. The former is an unresolved incoherent instrument; the latter is a coherent exchange-summed amplitude. Which description is physically appropriate depends on whether unobserved degrees of freedom retain which-pair information. The branch-individuation statement is common to both: if the record operator system is invariant under $S_2^{\left(e^{-}\right)}\times S_2^{\left(e^{+}\right)}$, then $(i,j)$ is not an admissible record label.

The no-jump operation over a short interval $\Delta t$ may be written schematically as

$${\mathcal{H}}_{⌀}\left(\rho \right)=M_0\rho M_0^{†},$$

(96)

with

$$M_0=I-i{H}_{\mathrm{eff}}\Delta t+O(\Delta t^2),$$

(97)

and

$$H_{\mathrm{eff}}=H-{\displaystyle \frac{i}{2}}\sum _{c\in \{2\gamma ,3\gamma \}}{\displaystyle \sum _{i,j=1}^2}{J}_{ij}^{c†}{J}_{ij}^c.$$

(98)

Thus the no-jump branch is also exchange-invariant: it is not a family of separately recorded events in which each artificial pair “failed” to annihilate. It is one no-jump record for the chosen monitoring interval.

If the detector resolves photon channel, the record-visible classes are

$$\{\left[h_{⌀}\right],\left[2\gamma \right],\left[3\gamma \right]\},N_{\mathrm{record}}=3.$$

(99)

If it records only annihilation versus no annihilation and not photon number, the two jump orbits are further coarse-grained:

$$\left[2\gamma \right]{\sim }_{\mathcal{A}}\left[3\gamma \right],$$

(100)

so that

$${\mathcal{B}}_{\mathrm{record}}=\{\left[h_{⌀}\right],\left[\mathrm{annihilation}\right]\},N_{\mathrm{record}}=2.$$

(101)

The hierarchy is therefore

$$N_{\mathrm{fine}}=9,N_{\mathrm{sym}}=3,N_{\mathrm{record}}=3\mathrm{or}2,$$

(102)

with the value of $N_{\mathrm{record}}$ determined by detector resolution. In the annihilation-only record algebra this is the reduction

$$9⟶3⟶2.$$

(103)

The lesson is not a new calculation of $P{s}_2$ decay. The formal content is the group action on labeled candidate histories, the induced exchange orbits, and the subsequent coarse-graining by a specified record operator system. Labeled pair indices are useful bookkeeping, but absent a physical record that distinguishes the annihilating electron and positron, they are not branch-individuating variables. This connects the present framework to exchange-summed no-jump survival in dipositronium [18] while keeping the dynamical assumptions explicit.

7. Partial Recoherence and Non-Monotone Effective Branching

Effective branch multiplicity need not be monotone. This point is secondary to the recovery theorem, but it prevents the packing number from being mistaken for a permanently growing tree count.

Consider

$$|\Psi \left(t\right)\rangle =c_0|0\rangle |E_0\left(t\right)\rangle +c_1|1\rangle |E_1\left(t\right)\rangle .$$

(104)

The environmental overlap is

$$\gamma \left(t\right)=\langle E_1\left(t\right)|E_0\left(t\right)\rangle .$$

(105)

When $\left|\gamma \right(t\left)\right|\approx 0$, the alternatives are effectively distinguishable relative to the system or pointer algebra. If $\left|\gamma \right(t\left)\right|$ later increases through partial recoherence, non-Markovian information backflow, or loss of environmental redundancy, the alternatives may cease to be $\epsilon$-distinguishable relative to $\mathcal{A}$.

Relative to a fixed algebra and resolution, one may have

$$N_{\mathrm{eff}}(t_1;\mathcal{A},\epsilon )=1,N_{\mathrm{eff}}(t_2;\mathcal{A},\epsilon )=2,N_{\mathrm{eff}}(t_3;\mathcal{A},\epsilon )=1.$$

(106)

This is not global world-merger. It is loss or recovery of effective distinguishability in the chosen algebra. The rise and fall of Darwinist redundancy provides one physical route to such behavior [9].

8. Relation to the Author’S Previous Work

The author’s previous work is relevant as motivation rather than as the formal basis of the argument. Exchange-summed no-jump survival in dipositronium motivates the warning that labeled constituent pairings need not be physical branch channels [18]. Common-mode dephasing in boson-fermion pairs motivates branch indistinguishability under collective monitoring [19]. Recoverability geometry in homogeneous and heterogeneous qudit decoherence-free structures provides examples where leakage shells and recovery maps determine which distinctions survive as logical distinctions [20]. These connections are illustrative of the three filters used here: symmetry, records, and recovery.

9. Discussion

The proposal can be compressed into one statement: branch multiplicity is an operational packing number on physically admissible history classes, after exact symmetries and algebra-relative distinguishability have specified what counts as the same or different branch.

The four-level hierarchy is

$$N_{\mathrm{fine}},N_{\mathrm{sym}},N_{\mathrm{record}},N_{\mathrm{logical}}.$$

(107)

A microscopic history may be distinct in a labeled bookkeeping space, identified by an exact symmetry, recorded in a syndrome, invisible to a pointer algebra, or absent at the recovered logical level. Proposition 1 makes the algebraic part of this hierarchy explicit: refining the algebra cannot reduce the number of distinguishable branch classes. Proposition 3 supplies the complementary channel statement: recovery and other physical post-processing cannot increase operational channel distinguishability. Theorem 4 then gives a cluster-count result: recovered logical multiplicity is the number of separated logical-channel clusters, not the number of microscopic histories assigned to those clusters.

The recovery theorem is intentionally modest. Its proof is the triangle inequality; its role is to fix the branch-individuation criterion. Histories that induce the same recovered logical channel are not distinct logical branches. That criterion imports the operational clarity of quantum error correction into branch individuation.

The collective-record theorem gives the packing formalism concrete force. A labeled exponential count can reduce to a polynomial exchange count and then to a linear collective-record count. The dipositronium example shows the same principle in channel language: if the record operator system does not resolve constituent labels, candidate histories are grouped by exchange orbits and then by detector records rather than by arbitrary labeled pair histories.

The framework does not solve the Born rule, does not supply an absolute number of worlds, and does not replace decoherence theory. It sharpens a prior step in branch-counting: before branches can be counted, one must specify the physical equivalence relation under which candidate histories are individuated.

10. Conclusion

Everettian branch multiplicity is not a raw count of wavefunction terms, nor simply a count of decoherent alternatives. It is an algebra-relative operational packing number of mutually distinguishable history classes, with orbit reduction, record compression, and recovery contraction as structural constraints.

Exact symmetries determine which labeled histories were never distinct physical alternatives. Record algebras determine which distinctions are physically visible. Recovery maps determine which microscopic histories survive as distinct logical branches. The hierarchy

$$N_{\mathrm{fine}},N_{\mathrm{sym}},N_{\mathrm{record}},N_{\mathrm{logical}}$$

(108)

therefore need not collapse to a single number.

A branch, in this framework, is not a labeled microscopic path. It is a symmetry-distinct, record-distinct, and, where relevant, logically distinct history class.