空間資訊佔據 IV:Generator Quotient 與跨表示資訊不變量
Spatial Information Occupancy IV: Generator Quotients and Cross-Representation Invariants
系列: Spatial Information Occupancy and Generator Geometry篇次: Paper 04 / Quotients and Representation版本: v1.0日期: 2026-10-02作者: Neo.KAI 協力研究者: Aletheia / ChatGPT, GPT-5.6 Sol機構: EveMissLab/一言諾科技有限公司狀態: Formal Research Draft / Exact Finite Quotient Theory
直接前置:
Paper 01《空間資訊佔據 I:超越幾何測度的容器相對自由結構》;
Paper 02《空間資訊佔據 II:自由度傳輸、邊界飽和與結構逃逸》;
Paper 03《空間資訊佔據 III:協同構成元代數、潛在自由與纖維資訊場》;
E4 Latent Freedom Potential and Irregular Cooperative Fiber Field;
E5 Generator Quotient;
主客邊三域代數(SOBTA);
Perfect Cuboid Global Closure 的 representation/world separation 方法論。
摘要
Paper 03 已建立 finite monotone freedom problem 的 minimal-generator representation:
H q = ( W , G q ) , \mathcal H_q
=
(W,\mathcal G_q), H q = ( W , G q ) ,
以及由此誘導的:
V q , d q , Φ q , J q . V_q,
\qquad
d_q,
\qquad
\Phi_q,
\qquad
\mathcal J_q. V q , d q , Φ q , J q .
但同一個 freedom structure 可以有大量不同表示:
constituent labels 可以重命名;
幾何 embedding 可以改變;
subset states 可以使用不同 radix 或 symbol encoding;
同一 hypergraph 可以有多個 incidence-matrix ordering;
rendering 可以改變,而 generator relations 不變。
因此若 representation 本身被當成底層資訊,就會把假的自由度誤認成真正結構。
本文正式定義:
X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) , \boxed{
X_1\sim_q X_2
\iff
\mathcal H_q(X_1)
\cong
\mathcal H_q(X_2),
} X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) ,
其中 hypergraph isomorphism 是一個 constituent 雙射:
φ : W 1 → W 2 \varphi:
W_1
\rightarrow
W_2 φ : W 1 → W 2
使:
G ∈ G 1 ⟺ φ ( G ) ∈ G 2 . G\in\mathcal G_1
\iff
\varphi(G)\in\mathcal G_2. G ∈ G 1 ⟺ φ ( G ) ∈ G 2 .
本文首先證明:
∼ q \boxed{
\sim_q
} ∼ q
是等價關係,因此 quotient:
Q q ( X ) = [ H q ( X ) ] ≅ \boxed{
\mathfrak Q_q(X)
=
[\mathcal H_q(X)]_{\cong}
} Q q ( X ) = [ H q ( X ) ] ≅
數學上良定義。
接著本文證明一組 isomorphism transport theorems 。若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則對所有:
A , B ⊆ W 1 , A,B\subseteq W_1, A , B ⊆ W 1 ,
有:
V 1 ( A ) = V 2 ( φ ( A ) ) , \boxed{
V_1(A)
=
V_2(\varphi(A)),
} V 1 ( A ) = V 2 ( φ ( A )) ,
d 1 ( A ) = d 2 ( φ ( A ) ) , \boxed{
d_1(A)
=
d_2(\varphi(A)),
} d 1 ( A ) = d 2 ( φ ( A )) ,
m 1 ( B ) = m 2 ( φ ( B ) ) , \boxed{
m_1(B)
=
m_2(\varphi(B)),
} m 1 ( B ) = m 2 ( φ ( B )) ,
以及 constituent Shapley field:
ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) . \boxed{
\phi_i^{(1)}
=
\phi_{\varphi(i)}^{(2)}.
} ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) .
因此 scalar / distributional quantities 可以形成 quotient invariants,而 vertex-indexed fields 應理解成 equivariant structures。
本文進一步證明:凡是由 hypergraph isomorphism class 唯一決定的 functional:
F ( H ) , F(\mathcal H), F ( H ) ,
皆可經 quotient factor:
F = F ~ ∘ π , \boxed{
F
=
\widetilde F
\circ
\pi,
} F = F ∘ π ,
其中:
π ( H ) = [ H ] ≅ . \pi(\mathcal H)
=
[\mathcal H]_{\cong}. π ( H ) = [ H ] ≅ .
這使 Generator Quotient 成為一個真正的 representation-independent carrier,而不是某種特定 hash。
對 finite-state encoding,本文再證明 Decoded Representation Invariance :若 encoder / decoder 構成雙射 recoding,則任何先 decode 再作用於底層 subset / hypergraph 的 quotient-compatible quantity皆保持不變。radix 2 , 3 , 8 , 10 , 16 , 36 2,3,8,10,16,36 2 , 3 , 8 , 10 , 16 , 36 只是其中一組實例。
最後,本文以四個互不重疊的三元 generators 作 baseline。其 automorphism group 大小為:
4 ! ( 3 ! ) 4 = 31104 , \boxed{
4!(3!)^4
=
31104,
} 4 ! ( 3 ! ) 4 = 31104 ,
所以在 12 12 12 個 labeled vertices 上,其 orbit 中 distinct labeled hypergraphs 數為:
12 ! 4 ! ( 3 ! ) 4 = 15400. \boxed{
\frac{12!}{4!(3!)^4}
=
15400.
} 4 ! ( 3 ! ) 4 12 ! = 15400.
也就是 15400 15400 15400 個不同 labeled manifestations 共享同一 Generator Quotient。
E5 的 negative control 使用含 6 6 6 個 size- 3 3 3 與 6 6 6 個 size- 4 4 4 generators 的 coupled model。因 edge-size multiset 已不同,它不可能與 baseline 同構,且其 winning-subset count 由 1695 1695 1695 變成 2131 2131 2131 。這證明 quotient 並非把所有 freedom systems 無差別壓成同一類。
本文的主要結論是:
representation can vary while the generator quotient remains fixed . \boxed{
\text{representation can vary while the generator quotient remains fixed}.
} representation can vary while the generator quotient remains fixed .
因此對指定 freedom query q q q ,Generator Quotient:
Q q \boxed{
\mathfrak Q_q
} Q q
構成第一個形式上明確的 cross-representation spatial-information invariant candidate。
Abstract
Paper 03 introduced the minimal-generator hypergraph
H q = ( W , G q ) \mathcal H_q=(W,\mathcal G_q) H q = ( W , G q )
for finite monotone freedom problems, together with the induced freedom predicate, latent-freedom potential, cooperative interactions, and constituent influence fields.
The same freedom structure may nevertheless admit many representations: vertex labels may be renamed, geometric embeddings may change, subset states may be encoded in different radices or symbolic alphabets, and incidence matrices may be reordered without changing the underlying generator relations.
This paper defines the equivalence relation
X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) X_1\sim_q X_2
\iff
\mathcal H_q(X_1)\cong\mathcal H_q(X_2) X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 )
and the corresponding Generator Quotient
Q q ( X ) = [ H q ( X ) ] ≅ . \mathfrak Q_q(X)
=
[\mathcal H_q(X)]_{\cong}. Q q ( X ) = [ H q ( X ) ] ≅ .
It proves that hypergraph isomorphisms transport the entire finite freedom algebra: the binary freedom predicate, latent-freedom potential, Möbius/Harsanyi coefficients, and Shapley fields satisfy exact invariant or equivariant laws under the induced subset bijection.
Any quantity depending only on the isomorphism class factors through the quotient. Conversely, vertex-indexed fields should be treated as equivariant associated structures rather than as label-invariant scalars.
A general decoded-representation theorem is also established: any bijective encoding with a correct inverse preserves quotient-compatible quantities after decoding. Cross-radix representations are therefore special cases of representation non-ownership rather than separate ontological structures.
A baseline hypergraph consisting of four disjoint triples has automorphism-group size
4 ! ( 3 ! ) 4 = 31104 4!(3!)^4=31104 4 ! ( 3 ! ) 4 = 31104
and an orbit of
15400 15400 15400
distinct labeled copies on twelve vertices. All belong to the same quotient class.
The main conclusion is that the generator quotient, rather than any particular labeling, radix, or embedding, is a natural finite candidate for cross-representation spatial-information invariance.
1. 問題:什麼才算真正改變?
Paper 03 已指出:
H q = ( W , G q ) \mathcal H_q
=
(W,\mathcal G_q) H q = ( W , G q )
足以重建 finite monotone freedom predicate:
V q . V_q. V q .
但若把 constituent:
1 , 2 , … , M 1,2,\ldots,M 1 , 2 , … , M
改名成:
a , b , … , a,b,\ldots, a , b , … ,
底層 freedom 不應因此被宣稱「改變」。
同樣,若 subset bitmask:
2730 2730 2730
寫成:
101010101010 2 101010101010_2 10101010101 0 2
或:
A A A 16 , AAA_{16}, AA A 16 ,
representation 不同,不代表 admissibility relation 改變。
所以本文追問:
哪些差異屬於 representation freedom,哪些差異屬於 generator structure 本身?
2. Freedom-Generator Hypergraph
令:
H = ( W , G ) \mathcal H
=
(W,\mathcal G) H = ( W , G )
其中:
W W W :finite constituent set;
G ⊆ 2 W \mathcal G\subseteq2^W G ⊆ 2 W :minimal freedom generator family。
本文假設:
G ≠ ∅ G\neq\varnothing G = ∅
對所有:
G ∈ G . G\in\mathcal G. G ∈ G .
binary freedom:
V H ( A ) = 1 [ ∃ G ∈ G : G ⊆ A ] . \boxed{
V_{\mathcal H}(A)
=
\mathbf 1
[
\exists G\in\mathcal G:
G\subseteq A
].
} V H ( A ) = 1 [ ∃ G ∈ G : G ⊆ A ] .
latent potential:
d H ( A ) = min G ∈ G ∣ G ∖ A ∣ . \boxed{
d_{\mathcal H}(A)
=
\min_{G\in\mathcal G}
|G\setminus A|.
} d H ( A ) = G ∈ G min ∣ G ∖ A ∣.
Möbius transform:
m H ( B ) = ∑ A ⊆ B ( − 1 ) ∣ B ∣ − ∣ A ∣ V H ( A ) . \boxed{
m_{\mathcal H}(B)
=
\sum_{A\subseteq B}
(-1)^{|B|-|A|}
V_{\mathcal H}(A).
} m H ( B ) = A ⊆ B ∑ ( − 1 ) ∣ B ∣ − ∣ A ∣ V H ( A ) .
Shapley field:
ϕ i ( H ) = ϕ i ( V H ) . \boxed{
\phi_i(\mathcal H)
=
\phi_i(V_{\mathcal H}).
} ϕ i ( H ) = ϕ i ( V H ) .
3. Hypergraph Isomorphism
Definition 3.1 — Generator Hypergraph Isomorphism
令:
H 1 = ( W 1 , G 1 ) , \mathcal H_1
=
(W_1,\mathcal G_1), H 1 = ( W 1 , G 1 ) ,
H 2 = ( W 2 , G 2 ) . \mathcal H_2
=
(W_2,\mathcal G_2). H 2 = ( W 2 , G 2 ) .
若存在雙射:
φ : W 1 → W 2 \boxed{
\varphi:
W_1
\rightarrow
W_2
} φ : W 1 → W 2
使:
G ∈ G 1 ⟺ φ ( G ) ∈ G 2 , \boxed{
G\in\mathcal G_1
\iff
\varphi(G)\in\mathcal G_2,
} G ∈ G 1 ⟺ φ ( G ) ∈ G 2 ,
其中:
φ ( G ) = { φ ( i ) : i ∈ G } , \varphi(G)
=
\{\varphi(i):i\in G\}, φ ( G ) = { φ ( i ) : i ∈ G } ,
則稱:
H 1 ≅ H 2 . \boxed{
\mathcal H_1
\cong
\mathcal H_2.
} H 1 ≅ H 2 .
φ \varphi φ 自然誘導 Boolean subset lattice 的雙射:
φ ^ : 2 W 1 → 2 W 2 , \boxed{
\widehat\varphi:
2^{W_1}
\rightarrow
2^{W_2},
} φ : 2 W 1 → 2 W 2 ,
φ ^ ( A ) = φ ( A ) . \widehat\varphi(A)
=
\varphi(A). φ ( A ) = φ ( A ) .
後文簡記為:
φ ( A ) . \varphi(A). φ ( A ) .
4. Generator Equivalence Relation
對同一 freedom query q q q 下的 admissible systems X X X ,定義:
X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) . \boxed{
X_1\sim_q X_2
\iff
\mathcal H_q(X_1)
\cong
\mathcal H_q(X_2).
} X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) .
Theorem 4.1 — ∼ q \sim_q ∼ q Is an Equivalence Relation
∼ q \sim_q ∼ q 滿足:
reflexivity;
symmetry;
transitivity。
Proof
Reflexivity
identity map:
id W : W → W \operatorname{id}_W:
W\rightarrow W id W : W → W
顯然保持所有 generators,因此:
H ≅ H . \mathcal H\cong\mathcal H. H ≅ H .
Symmetry
若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則 φ \varphi φ 為雙射,因此:
φ − 1 \varphi^{-1} φ − 1
存在。
generator-preserving 條件立即推出:
φ − 1 : H 2 ≅ H 1 . \varphi^{-1}:
\mathcal H_2
\cong
\mathcal H_1. φ − 1 : H 2 ≅ H 1 .
Transitivity
若:
φ : H 1 ≅ H 2 \varphi:
\mathcal H_1
\cong
\mathcal H_2 φ : H 1 ≅ H 2
且:
ψ : H 2 ≅ H 3 , \psi:
\mathcal H_2
\cong
\mathcal H_3, ψ : H 2 ≅ H 3 ,
則:
ψ ∘ φ \psi\circ\varphi ψ ∘ φ
為雙射,且把:
G 1 \mathcal G_1 G 1
依序搬到:
G 2 \mathcal G_2 G 2
再搬到:
G 3 . \mathcal G_3. G 3 .
因此:
ψ ∘ φ : H 1 ≅ H 3 . \psi\circ\varphi:
\mathcal H_1
\cong
\mathcal H_3. ψ ∘ φ : H 1 ≅ H 3 .
證畢。
5. Generator Quotient
由 Theorem 4.1,可正式定義:
Definition 5.1 — Generator Quotient
Q q ( X ) = [ H q ( X ) ] ≅ . \boxed{
\mathfrak Q_q(X)
=
[
\mathcal H_q(X)
]_{\cong}.
} Q q ( X ) = [ H q ( X ) ] ≅ .
這是一個 isomorphism class,而不是:
vertex label list;
incidence-matrix row ordering;
radix string;
geometric rendering;
hash digest。
因此:
Q q \boxed{
\mathfrak Q_q
} Q q
是 abstract relational object。
6. Freedom Predicate Equivariance
Theorem 6.1 — State Equivariance
若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則對所有:
A ⊆ W 1 , A\subseteq W_1, A ⊆ W 1 ,
有:
V H 1 ( A ) = V H 2 ( φ ( A ) ) . \boxed{
V_{\mathcal H_1}(A)
=
V_{\mathcal H_2}(\varphi(A)).
} V H 1 ( A ) = V H 2 ( φ ( A )) .
Proof
V H 1 ( A ) = 1 V_{\mathcal H_1}(A)=1 V H 1 ( A ) = 1
當且僅當存在:
G ∈ G 1 G\in\mathcal G_1 G ∈ G 1
使:
G ⊆ A . G\subseteq A. G ⊆ A .
因 φ \varphi φ 為雙射:
G ⊆ A ⟺ φ ( G ) ⊆ φ ( A ) . G\subseteq A
\iff
\varphi(G)\subseteq\varphi(A). G ⊆ A ⟺ φ ( G ) ⊆ φ ( A ) .
又因 hypergraph isomorphism:
G ∈ G 1 ⟺ φ ( G ) ∈ G 2 . G\in\mathcal G_1
\iff
\varphi(G)\in\mathcal G_2. G ∈ G 1 ⟺ φ ( G ) ∈ G 2 .
所以:
V H 1 ( A ) = 1 ⟺ V H 2 ( φ ( A ) ) = 1. V_{\mathcal H_1}(A)=1
\iff
V_{\mathcal H_2}(\varphi(A))=1. V H 1 ( A ) = 1 ⟺ V H 2 ( φ ( A )) = 1.
binary value 相等。證畢。
7. Latent Potential Equivariance
Theorem 7.1
若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則:
d H 1 ( A ) = d H 2 ( φ ( A ) ) \boxed{
d_{\mathcal H_1}(A)
=
d_{\mathcal H_2}(\varphi(A))
} d H 1 ( A ) = d H 2 ( φ ( A ))
對所有:
A ⊆ W 1 A\subseteq W_1 A ⊆ W 1
成立。
Proof
由定義:
d H 2 ( φ ( A ) ) = min G ′ ∈ G 2 ∣ G ′ ∖ φ ( A ) ∣ . d_{\mathcal H_2}(\varphi(A))
=
\min_{G'\in\mathcal G_2}
|G'\setminus\varphi(A)|. d H 2 ( φ ( A )) = G ′ ∈ G 2 min ∣ G ′ ∖ φ ( A ) ∣.
每個:
G ′ ∈ G 2 G'\in\mathcal G_2 G ′ ∈ G 2
唯一寫成:
G ′ = φ ( G ) , G ∈ G 1 . G'=\varphi(G),
\qquad
G\in\mathcal G_1. G ′ = φ ( G ) , G ∈ G 1 .
且雙射保持 set difference cardinality:
∣ φ ( G ) ∖ φ ( A ) ∣ = ∣ φ ( G ∖ A ) ∣ = ∣ G ∖ A ∣ . |\varphi(G)\setminus\varphi(A)|
=
|\varphi(G\setminus A)|
=
|G\setminus A|. ∣ φ ( G ) ∖ φ ( A ) ∣ = ∣ φ ( G ∖ A ) ∣ = ∣ G ∖ A ∣.
所以:
d H 2 ( φ ( A ) ) = min G ∈ G 1 ∣ G ∖ A ∣ = d H 1 ( A ) . d_{\mathcal H_2}(\varphi(A))
=
\min_{G\in\mathcal G_1}
|G\setminus A|
=
d_{\mathcal H_1}(A). d H 2 ( φ ( A )) = G ∈ G 1 min ∣ G ∖ A ∣ = d H 1 ( A ) .
證畢。
8. Möbius Equivariance
Theorem 8.1
若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則:
m H 1 ( B ) = m H 2 ( φ ( B ) ) \boxed{
m_{\mathcal H_1}(B)
=
m_{\mathcal H_2}(\varphi(B))
} m H 1 ( B ) = m H 2 ( φ ( B ))
對所有:
B ⊆ W 1 B\subseteq W_1 B ⊆ W 1
成立。
Proof
由 Möbius 定義:
m H 2 ( φ ( B ) ) = ∑ C ⊆ φ ( B ) ( − 1 ) ∣ φ ( B ) ∣ − ∣ C ∣ V H 2 ( C ) . m_{\mathcal H_2}(\varphi(B))
=
\sum_{C\subseteq\varphi(B)}
(-1)^{|\varphi(B)|-|C|}
V_{\mathcal H_2}(C). m H 2 ( φ ( B )) = C ⊆ φ ( B ) ∑ ( − 1 ) ∣ φ ( B ) ∣ − ∣ C ∣ V H 2 ( C ) .
因 φ \varphi φ 在 subset lattice 上是雙射,所有:
C ⊆ φ ( B ) C\subseteq\varphi(B) C ⊆ φ ( B )
唯一表示為:
C = φ ( A ) , A ⊆ B . C=\varphi(A),
\qquad
A\subseteq B. C = φ ( A ) , A ⊆ B .
又:
∣ φ ( B ) ∣ = ∣ B ∣ , |\varphi(B)|=|B|, ∣ φ ( B ) ∣ = ∣ B ∣ ,
∣ φ ( A ) ∣ = ∣ A ∣ . |\varphi(A)|=|A|. ∣ φ ( A ) ∣ = ∣ A ∣.
由 Theorem 6.1:
V H 2 ( φ ( A ) ) = V H 1 ( A ) . V_{\mathcal H_2}(\varphi(A))
=
V_{\mathcal H_1}(A). V H 2 ( φ ( A )) = V H 1 ( A ) .
所以:
m H 2 ( φ ( B ) ) = ∑ A ⊆ B ( − 1 ) ∣ B ∣ − ∣ A ∣ V H 1 ( A ) = m H 1 ( B ) . \begin{aligned}
m_{\mathcal H_2}(\varphi(B))
&=
\sum_{A\subseteq B}
(-1)^{|B|-|A|}
V_{\mathcal H_1}(A)\\
&=
m_{\mathcal H_1}(B).
\end{aligned} m H 2 ( φ ( B )) = A ⊆ B ∑ ( − 1 ) ∣ B ∣ − ∣ A ∣ V H 1 ( A ) = m H 1 ( B ) .
證畢。
9. Shapley Equivariance
Theorem 9.1
若:
φ : H 1 ≅ H 2 , \varphi:
\mathcal H_1
\cong
\mathcal H_2, φ : H 1 ≅ H 2 ,
則對每個:
i ∈ W 1 , i\in W_1, i ∈ W 1 ,
有:
ϕ i ( H 1 ) = ϕ φ ( i ) ( H 2 ) . \boxed{
\phi_i(\mathcal H_1)
=
\phi_{\varphi(i)}(\mathcal H_2).
} ϕ i ( H 1 ) = ϕ φ ( i ) ( H 2 ) .
Proof
Shapley value:
ϕ i ( H 1 ) = ∑ A ⊆ W 1 ∖ { i } ∣ A ∣ ! ( n − ∣ A ∣ − 1 ) ! n ! [ V 1 ( A ∪ { i } ) − V 1 ( A ) ] , \phi_i(\mathcal H_1)
=
\sum_{A\subseteq W_1\setminus\{i\}}
\frac{
|A|!(n-|A|-1)!
}{
n!
}
\left[
V_1(A\cup\{i\})-V_1(A)
\right], ϕ i ( H 1 ) = A ⊆ W 1 ∖ { i } ∑ n ! ∣ A ∣ ! ( n − ∣ A ∣ − 1 )! [ V 1 ( A ∪ { i }) − V 1 ( A ) ] ,
其中:
n = ∣ W 1 ∣ = ∣ W 2 ∣ . n=|W_1|=|W_2|. n = ∣ W 1 ∣ = ∣ W 2 ∣.
φ \varphi φ 把:
W 1 ∖ { i } W_1\setminus\{i\} W 1 ∖ { i }
的 subset lattice 雙射到:
W 2 ∖ { φ ( i ) } . W_2\setminus\{\varphi(i)\}. W 2 ∖ { φ ( i )} .
且:
∣ φ ( A ) ∣ = ∣ A ∣ . |\varphi(A)|=|A|. ∣ φ ( A ) ∣ = ∣ A ∣.
由 Theorem 6.1:
V 1 ( A ) = V 2 ( φ ( A ) ) , V_1(A)
=
V_2(\varphi(A)), V 1 ( A ) = V 2 ( φ ( A )) ,
以及:
V 1 ( A ∪ { i } ) = V 2 ( φ ( A ) ∪ { φ ( i ) } ) . V_1(A\cup\{i\})
=
V_2(
\varphi(A)\cup\{\varphi(i)\}
). V 1 ( A ∪ { i }) = V 2 ( φ ( A ) ∪ { φ ( i )}) .
所以逐項權重與 marginal contribution 完全一致。
因此:
ϕ i ( H 1 ) = ϕ φ ( i ) ( H 2 ) . \phi_i(\mathcal H_1)
=
\phi_{\varphi(i)}(\mathcal H_2). ϕ i ( H 1 ) = ϕ φ ( i ) ( H 2 ) .
證畢。
10. Invariant 與 Equivariant 必須分開
Theorems 6.1–9.1 顯示不同 quantity 有不同 transformation law。
10.1 Isomorphism-invariant quantities
例如:
∣ W ∣ |W| ∣ W ∣ ;
generator edge-size multiset;
winning-subset count;
winning counts by subset cardinality;
latent-distance histogram;
Möbius order/coefficient spectrum;
sorted Shapley multiset;
automorphism-group size。
這些不需要保留 vertex correspondence。
10.2 Equivariant quantities
例如:
A ↦ V ( A ) , A
\mapsto
V(A), A ↦ V ( A ) ,
A ↦ d ( A ) , A
\mapsto
d(A), A ↦ d ( A ) ,
B ↦ m ( B ) , B
\mapsto
m(B), B ↦ m ( B ) ,
i ↦ ϕ i . i
\mapsto
\phi_i. i ↦ ϕ i .
它們不是在「相同 label」下保持,而是在:
φ \varphi φ
搬移後保持。
因此:
Invariant ≠ Equivariant . \boxed{
\text{Invariant}
\neq
\text{Equivariant}.
} Invariant = Equivariant .
這是 cross-representation theory 的必要區分。
11. Quotient Factorization
令:
H \mathfrak H H
為某一類 finite freedom-generator hypergraphs 的集合。
定義 quotient projection:
π : H → H / ≅ , \boxed{
\pi:
\mathfrak H
\rightarrow
\mathfrak H/{\cong},
} π : H → H / ≅ ,
π ( H ) = [ H ] ≅ . \pi(\mathcal H)
=
[\mathcal H]_{\cong}. π ( H ) = [ H ] ≅ .
Theorem 11.1 — Factorization Through Generator Quotient
若 functional:
F : H → Y F:
\mathfrak H
\rightarrow
Y F : H → Y
滿足:
H 1 ≅ H 2 ⇒ F ( H 1 ) = F ( H 2 ) , \mathcal H_1\cong\mathcal H_2
\Rightarrow
F(\mathcal H_1)=F(\mathcal H_2), H 1 ≅ H 2 ⇒ F ( H 1 ) = F ( H 2 ) ,
則存在唯一:
F ~ : H / ≅ → Y \widetilde F:
\mathfrak H/{\cong}
\rightarrow
Y F : H / ≅ → Y
使:
F = F ~ ∘ π . \boxed{
F
=
\widetilde F
\circ
\pi.
} F = F ∘ π .
Proof
對 quotient class:
[ H ] ≅ , [\mathcal H]_{\cong}, [ H ] ≅ ,
定義:
F ~ ( [ H ] ≅ ) = F ( H ) . \widetilde F(
[\mathcal H]_{\cong}
)
=
F(\mathcal H). F ([ H ] ≅ ) = F ( H ) .
因 F F F 對 isomorphism invariant,此定義與 representative 無關,因此 well-defined。
且:
( F ~ ∘ π ) ( H ) = F ~ ( [ H ] ) = F ( H ) . (\widetilde F\circ\pi)(\mathcal H)
=
\widetilde F([\mathcal H])
=
F(\mathcal H). ( F ∘ π ) ( H ) = F ([ H ]) = F ( H ) .
唯一性由:
π \pi π
surjective 得到。證畢。
12. Generator Quotient 是什麼,不是什麼
Q q = [ H q ] ≅ \boxed{
\mathfrak Q_q
=
[\mathcal H_q]_{\cong}
} Q q = [ H q ] ≅
不是一個 scalar。
不是:
area;
entropy number;
Shapley list;
hash;
radix-independent integer。
它是一個 equivalence class of relational structures。
後續所有 scalar fingerprints 都只是:
functions on the quotient . \boxed{
\text{functions on the quotient}.
} functions on the quotient .
因此不應把 fingerprint 與 quotient 本身混同。
13. Fingerprints
為計算與快速比較,可定義 isomorphism-invariant fingerprints。
例如:
Definition 13.1 — Winning-Count Profile
W k ( H ) = # { A ⊆ W : ∣ A ∣ = k , V H ( A ) = 1 } . \boxed{
W_k(\mathcal H)
=
\#\{
A\subseteq W:
|A|=k,
V_{\mathcal H}(A)=1
\}.
} W k ( H ) = # { A ⊆ W : ∣ A ∣ = k , V H ( A ) = 1 } .
形成:
W = ( W 0 , … , W n ) . \boxed{
\mathbf W
=
(W_0,\ldots,W_n).
} W = ( W 0 , … , W n ) .
Definition 13.2 — Latent-Distance Histogram
D k , r = # { A : ∣ A ∣ = k , d H ( A ) = r } . \boxed{
D_{k,r}
=
\#\{
A:
|A|=k,
d_{\mathcal H}(A)=r
\}.
} D k , r = # { A : ∣ A ∣ = k , d H ( A ) = r } .
Definition 13.3 — Möbius Spectrum
記錄:
( ∣ B ∣ , m H ( B ) ) \boxed{
(|B|,m_{\mathcal H}(B))
} ( ∣ B ∣ , m H ( B ))
的 multiplicities。
Definition 13.4 — Shapley Multiset
{ ϕ i : i ∈ W } . \boxed{
\{
\phi_i
:
i\in W
\}.
} { ϕ i : i ∈ W } .
Theorems 6.1–9.1 立即推出這些 quantities 在 hypergraph isomorphism 下 invariant。
14. Fingerprint Incompleteness
本文特別不宣稱:
same finite fingerprint ⇒ hypergraph isomorphism . \boxed{
\text{same finite fingerprint}
\Rightarrow
\text{hypergraph isomorphism}.
} same finite fingerprint ⇒ hypergraph isomorphism .
一般而言,不同非同構結構可以共享某些:
degree sequences;
edge-size multisets;
spectra;
aggregate counts。
因此:
fingerprint ≠ canonical quotient object . \boxed{
\text{fingerprint}
\neq
\text{canonical quotient object}.
} fingerprint = canonical quotient object .
若需要 exact equality of Generator Quotients,應進行:
hypergraph isomorphism test \boxed{
\text{hypergraph isomorphism test}
} hypergraph isomorphism test
或使用已證完整的 canonical form。
15. Representation Map
現在把 representation layer 明確加入。
令底層 finite state space:
Ω . \Omega. Ω.
一個 representation system 為:
ρ : Ω → Σ \boxed{
\rho:
\Omega
\rightarrow
\Sigma
} ρ : Ω → Σ
與 decoder:
δ : ρ ( Ω ) → Ω . \boxed{
\delta:
\rho(\Omega)
\rightarrow
\Omega.
} δ : ρ ( Ω ) → Ω.
若:
δ ∘ ρ = id Ω , \delta\circ\rho
=
\operatorname{id}_\Omega, δ ∘ ρ = id Ω ,
則 representation 是 lossless on Ω \Omega Ω 。
16. Decoded Representation Invariance
Theorem 16.1
令:
I : Ω → Y I:
\Omega
\rightarrow
Y I : Ω → Y
為任意底層 quantity。
若:
ρ : Ω → Σ \rho:
\Omega\rightarrow\Sigma ρ : Ω → Σ
具有左逆 decoder:
δ \delta δ
使:
δ ( ρ ( ω ) ) = ω , \delta(\rho(\omega))=\omega, δ ( ρ ( ω )) = ω ,
則定義 represented quantity:
I ρ ( s ) = I ( δ ( s ) ) . I_{\rho}(s)
=
I(\delta(s)). I ρ ( s ) = I ( δ ( s )) .
對所有:
ω ∈ Ω , \omega\in\Omega, ω ∈ Ω ,
有:
I ρ ( ρ ( ω ) ) = I ( ω ) . \boxed{
I_{\rho}(\rho(\omega))
=
I(\omega).
} I ρ ( ρ ( ω )) = I ( ω ) .
Proof
I ρ ( ρ ( ω ) ) = I ( δ ( ρ ( ω ) ) ) = I ( ω ) . I_{\rho}(\rho(\omega))
=
I(
\delta(\rho(\omega))
)
=
I(\omega). I ρ ( ρ ( ω )) = I ( δ ( ρ ( ω ))) = I ( ω ) .
證畢。
Corollary 16.2 — Radix Non-Ownership
若 finite subset state 先被 bijectively 編成 integer:
m ( A ) , m(A), m ( A ) ,
再用 base:
b b b
轉成 digit string:
ρ b ( A ) , \rho_b(A), ρ b ( A ) ,
且 decoder 可無損回復:
A , A, A ,
則任何對底層 subset / hypergraph 定義的 quantity:
I ( A ) I(A) I ( A )
都滿足:
I ( decode b ( ρ b ( A ) ) ) = I ( A ) . \boxed{
I(
\operatorname{decode}_b(\rho_b(A))
)
=
I(A).
} I ( decode b ( ρ b ( A ))) = I ( A ) .
因此:
radix does not own the decoded relational invariant . \boxed{
\text{radix does not own the decoded relational invariant}.
} radix does not own the decoded relational invariant .
17. 什麼可以隨 representation 改變?
Theorem 16.1 不表示所有東西都 invariant。
以下可以 representation-dependent:
raw string length;
digit frequency;
carry pattern;
byte length;
parser complexity;
local textual regularity;
compression ratio;
hardware execution cost。
因此應區分:
Representation-Owned Quantity \boxed{
\text{Representation-Owned Quantity}
} Representation-Owned Quantity
與:
Decoded Relational Quantity . \boxed{
\text{Decoded Relational Quantity}.
} Decoded Relational Quantity .
兩者都可能有研究價值,但不能混為同一層。
18. Baseline Example:四個互不重疊的三元 Generators
令:
W = { 0 , … , 11 } , W
=
\{0,\ldots,11\}, W = { 0 , … , 11 } ,
以及:
G = { G 1 , G 2 , G 3 , G 4 } , \mathcal G
=
\{
G_1,G_2,G_3,G_4
\}, G = { G 1 , G 2 , G 3 , G 4 } ,
其中:
G 1 = { 0 , 1 , 2 } , G_1=\{0,1,2\}, G 1 = { 0 , 1 , 2 } ,
G 2 = { 3 , 4 , 5 } , G_2=\{3,4,5\}, G 2 = { 3 , 4 , 5 } ,
G 3 = { 6 , 7 , 8 } , G_3=\{6,7,8\}, G 3 = { 6 , 7 , 8 } ,
G 4 = { 9 , 10 , 11 } . G_4=\{9,10,11\}. G 4 = { 9 , 10 , 11 } .
19. Automorphism Group
任何 automorphism 可以:
任意排列四個 triple blocks;
在每個 triple 內任意排列三個 vertices。
所以:
Aut ( H ) ≅ ( S 3 ) 4 ⋊ S 4 . \operatorname{Aut}(\mathcal H)
\cong
(S_3)^4
\rtimes
S_4. Aut ( H ) ≅ ( S 3 ) 4 ⋊ S 4 .
Theorem 19.1 — Baseline Automorphism-Group Size
∣ Aut ( H ) ∣ = 4 ! ( 3 ! ) 4 = 31104. \boxed{
|\operatorname{Aut}(\mathcal H)|
=
4!(3!)^4
=
31104.
} ∣ Aut ( H ) ∣ = 4 ! ( 3 ! ) 4 = 31104.
Proof
四個 blocks 可有:
4 ! 4! 4 !
種 permutation。
每個 block 內有:
3 ! 3! 3 !
種 vertex permutation,四個 blocks 獨立,所以:
( 3 ! ) 4 . (3!)^4. ( 3 ! ) 4 .
相乘:
4 ! ( 3 ! ) 4 . 4!(3!)^4. 4 ! ( 3 ! ) 4 .
任何 automorphism 必須把 size- 3 3 3 connected generator block 搬到另一個 generator block,且在 block 內任意 permutation;不存在額外操作。
證畢。
20. Labeled Orbit Size
對 symmetric group:
S 12 S_{12} S 12
作用於所有 labeled copies。
由 orbit-stabilizer:
∣ Orb ( H ) ∣ = ∣ S 12 ∣ ∣ Aut ( H ) ∣ . |\operatorname{Orb}(\mathcal H)|
=
\frac{
|S_{12}|
}{
|\operatorname{Aut}(\mathcal H)|
}. ∣ Orb ( H ) ∣ = ∣ Aut ( H ) ∣ ∣ S 12 ∣ .
Corollary 20.1
∣ Orb ( H ) ∣ = 12 ! 4 ! ( 3 ! ) 4 = 15400. \boxed{
|\operatorname{Orb}(\mathcal H)|
=
\frac{
12!
}{
4!(3!)^4
}
=
15400.
} ∣ Orb ( H ) ∣ = 4 ! ( 3 ! ) 4 12 ! = 15400.
因此同一 Generator Quotient 在固定 12 12 12 -vertex label universe 上至少有:
15400 15400 15400
個 distinct labeled manifestations。
這精確展示:
representation multiplicity ≫ 1. \boxed{
\text{representation multiplicity}
\gg
1.
} representation multiplicity ≫ 1.
21. State-Space Consequence
baseline binary freedom:
V ( A ) = 1 V(A)
=
1 V ( A ) = 1
當且僅當:
A A A
完整包含至少一個 triple generator。
總 subsets:
2 12 = 4096. 2^{12}
=
4096. 2 12 = 4096.
由 inclusion-exclusion:
N w i n = 4 ⋅ 2 9 − ( 4 2 ) 2 6 + ( 4 3 ) 2 3 − 1 = 2048 − 384 + 32 − 1 = 1695. \begin{aligned}
N_{\mathrm{win}}
&=
4\cdot2^9
-
\binom42 2^6
+
\binom43 2^3
-
1\\
&=
2048-384+32-1\\
&=
\boxed{
1695.
}
\end{aligned} N win = 4 ⋅ 2 9 − ( 2 4 ) 2 6 + ( 3 4 ) 2 3 − 1 = 2048 − 384 + 32 − 1 = 1695.
所以:
1695 \boxed{
1695
} 1695
本身也是 quotient invariant。
22. E5 Cross-Radix Exact Audit
E5 對全部:
4096 4096 4096
個 subset states 使用:
b ∈ { 2 , 3 , 8 , 10 , 16 , 36 } b\in
\{2,3,8,10,16,36\} b ∈ { 2 , 3 , 8 , 10 , 16 , 36 }
做 lossless encode / decode。
代表 state:
m = 2730 m=2730 m = 2730
呈現為:
base 2 2 2 :101010101010;
base 3 3 3 :10202010;
base 8 8 8 :5252;
base 10 10 10 :2730;
base 16 16 16 :AAA;
base 36 36 36 :23U。
raw strings 全不同。
但 decode 後的:
V ( A ) , d ( A ) , m ( A ) V(A),
\qquad
d(A),
\qquad
m(A) V ( A ) , d ( A ) , m ( A )
full-state audit digest 全部一致。
這是 Theorem 16.1 的 finite computational audit,而不是其 proof source。
23. Negative Control
E4 / E5 coupled model 有:
12 12 12
個 minimal generators:
6 6 6 個 size- 3 3 3 ;
6 6 6 個 size- 4 4 4 。
baseline 則只有:
4 4 4
個 size- 3 3 3 generators。
Proposition 23.1
兩者不可能 hypergraph-isomorphic。
Proof
hypergraph isomorphism 保持 edge cardinalities 的 multiset。
baseline edge-size multiset:
{ 3 , 3 , 3 , 3 } . \{3,3,3,3\}. { 3 , 3 , 3 , 3 } .
coupled edge-size multiset:
{ 3 , 3 , 3 , 3 , 3 , 3 , 4 , 4 , 4 , 4 , 4 , 4 } . \{
3,3,3,3,3,3,
4,4,4,4,4,4
\}. { 3 , 3 , 3 , 3 , 3 , 3 , 4 , 4 , 4 , 4 , 4 , 4 } .
兩者不同,所以不存在 isomorphism。證畢。
Computational Consequence
E5 exact enumeration:
baseline winning subsets:
1695 , 1695, 1695 ,
coupled winning subsets:
2131. 2131. 2131.
因此某些 quotient-compatible fingerprints 也立即分開。
這表示 Generator Quotient 並沒有把所有 freedom systems 都壓成同一類。
24. Embedding Redundancy
E4 還提供一個更具幾何意義的例子。
straight-separated 與 zigzag-separated:
barrier support 都是 72 72 72 ;
candidate constituents 都是 12 12 12 ;
interface edge count 從 54 54 54 變成 76 76 76 ;
bbox width 從 3 3 3 變成 6 6 6 。
幾何 embedding 明顯改變。
但兩者 minimal-generator hypergraphs 同構,且在採用一致 abstract labels 時甚至完全相同。
因此:
geometric embedding change ⇏ Generator Quotient change . \boxed{
\text{geometric embedding change}
\not\Rightarrow
\text{Generator Quotient change}.
} geometric embedding change ⇒ Generator Quotient change .
這不表示 geometry 不重要。
它表示:
對指定 freedom query,geometry 只有在它改變 admissibility generator relations 時,才改變 Q q \mathfrak Q_q Q q 。
25. Generator Quotient as a Candidate Information Invariant
因此本文提出:
Q q ( X ) = [ H q ( X ) ] ≅ \boxed{
\mathfrak Q_q(X)
=
[
\mathcal H_q(X)
]_{\cong}
} Q q ( X ) = [ H q ( X ) ] ≅
作為 finite cross-representation spatial-information invariant candidate。
其優點:
不依賴 constituent names;
不依賴 incidence ordering;
不依賴 radix rendering;
可以容納多種 geometric embeddings;
保留 generator relations;
能導出 binary freedom、latent distance、Möbius、Shapley 等結構。
但其限制同樣明確:
Q q \boxed{
\mathfrak Q_q
} Q q
仍然依賴:
query q q q ;
constituent resolution;
admissibility semantics;
container model。
因此它不是 universal absolute invariant。
Paper 05 將處理 constituent count 改變時的 cross-resolution 問題。
26. SOBTA:No Single Domain Owns the Quotient
對:
Γ q = ⟨ S q , O , ∂ ⟩ , \Gamma_q
=
\langle
S_q,O,\partial\rangle, Γ q = ⟨ S q , O , ∂ ⟩ ,
generator hypergraph 可寫:
H q = Gen ( S q , O , ∂ ) . \mathcal H_q
=
\operatorname{Gen}
(
S_q,O,\partial
). H q = Gen ( S q , O , ∂ ) .
所以 quotient:
Q q \mathfrak Q_q Q q
也不是單獨由:
S q , S_q, S q ,
O , O, O ,
或:
∂ \partial ∂
任何一域擁有。
它是三者在指定 admissibility semantics 下共同誘導的 relational structure。
這呼應 SOBTA:
No Single-Domain Ownership . \boxed{
\text{No Single-Domain Ownership}.
} No Single-Domain Ownership .
27. Projection without Annexation
若兩個幾何 systems:
X 1 , X 2 X_1,
X_2 X 1 , X 2
有:
Q q ( X 1 ) = Q q ( X 2 ) , \mathfrak Q_q(X_1)
=
\mathfrak Q_q(X_2), Q q ( X 1 ) = Q q ( X 2 ) ,
只能推出:
對 query q q q ,它們具有同構 generator structure。
不能推出:
X 1 = X 2 \boxed{
X_1=X_2
} X 1 = X 2
作為完整幾何/物理本體。
因此:
Generator equivalence ≠ ontological identity . \boxed{
\text{Generator equivalence}
\neq
\text{ontological identity}.
} Generator equivalence = ontological identity .
這就是本系列對 SOBTA:
Projection without Annexation \text{Projection without Annexation} Projection without Annexation
的直接實踐。
28. ACTM:Quotient Transport
現在 operator:
T : X t → X t + 1 T:
X_t
\rightarrow
X_{t+1} T : X t → X t + 1
可以被提升成:
Q q ( X t ) → T Q q ( X t + 1 ) . \boxed{
\mathfrak Q_q(X_t)
\xrightarrow{T}
\mathfrak Q_q(X_{t+1}).
} Q q ( X t ) T Q q ( X t + 1 ) .
可分:
Representation-Preserving
labels / encoding 改變,但:
Q q ( X t ) = Q q ( X t + 1 ) . \mathfrak Q_q(X_t)
=
\mathfrak Q_q(X_{t+1}). Q q ( X t ) = Q q ( X t + 1 ) .
Embedding-Changing, Quotient-Preserving
幾何 rendering 改變,但 generator hypergraph 仍同構。
Quotient-Changing
Q q ( X t ) ≠ Q q ( X t + 1 ) . \boxed{
\mathfrak Q_q(X_t)
\neq
\mathfrak Q_q(X_{t+1}).
} Q q ( X t ) = Q q ( X t + 1 ) .
最後一種才是 finite generator structure 的真正 structural transition。
29. 可證偽性與限制
F1 — Quotient Ill-Defined
若 ∼ q \sim_q ∼ q 不是等價關係,quotient 失效。
Theorem 4.1 已解析排除。
F2 — State Non-Equivariance
若 hypergraph isomorphism 下 binary freedom 不保持,Theorem 6.1 失效。
F3 — Latent / Möbius / Shapley Non-Equivariance
若任一 derived structure 依賴 arbitrary labels,Theorems 7.1–9.1 失效。
F4 — Encoding Confusion
如果 representation 是 lossy:
δ ∘ ρ ≠ id , \delta\circ\rho
\neq
\operatorname{id}, δ ∘ ρ = id ,
Theorem 16.1 不適用。
因此:
lossless recoding \boxed{
\text{lossless recoding}
} lossless recoding
與:
lossy coarse-graining \boxed{
\text{lossy coarse-graining}
} lossy coarse-graining
必須分開。
Paper 05 將專門處理後者。
F5 — Fingerprint Collision
aggregate fingerprint 一致不能自動當成 isomorphism proof。
F6 — Query Dependence
同一 geometry 在:
q 1 q_1 q 1
與:
q 2 q_2 q 2
下可有不同 quotient:
Q q 1 ≠ Q q 2 . \mathfrak Q_{q_1}
\neq
\mathfrak Q_{q_2}. Q q 1 = Q q 2 .
因此 invariant 是 query-indexed,而不是絕對無條件。
30. 與下一篇的接口
Paper 04 解決的是:
vertex 數不變時,如何 quotient 掉 representation redundancy?
但解析度變化:
N → 2 N N
\rightarrow
2N N → 2 N
時:
∣ W N ∣ ≠ ∣ W 2 N ∣ . |W_N|
\neq
|W_{2N}|. ∣ W N ∣ = ∣ W 2 N ∣.
此時普通 hypergraph isomorphism不可能成立。
所以 Paper 05 將研究:
{ H q , N } N \boxed{
\{
\mathcal H_{q,N}
\}_N
} { H q , N } N
作為 refinement system,並導入:
coarse / fine morphism;
inner / outer bracketing;
generator-start Hausdorff limit;
overlap kernel;
latent-potential normalization;
constituent field convergence。
問題將從:
representation invariance \boxed{
\text{representation invariance}
} representation invariance
提升到:
substrate-resolution invariance . \boxed{
\text{substrate-resolution invariance}.
} substrate-resolution invariance .
31. 結論
本文正式建立 Generator Quotient:
Q q ( X ) = [ H q ( X ) ] ≅ . \boxed{
\mathfrak Q_q(X)
=
[
\mathcal H_q(X)
]_{\cong}.
} Q q ( X ) = [ H q ( X ) ] ≅ .
其基礎等價關係:
X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 ) X_1\sim_q X_2
\iff
\mathcal H_q(X_1)
\cong
\mathcal H_q(X_2) X 1 ∼ q X 2 ⟺ H q ( X 1 ) ≅ H q ( X 2 )
已證明為真正 equivalence relation。
在 isomorphism:
φ \varphi φ
下:
V 1 ( A ) = V 2 ( φ ( A ) ) , \boxed{
V_1(A)
=
V_2(\varphi(A)),
} V 1 ( A ) = V 2 ( φ ( A )) ,
d 1 ( A ) = d 2 ( φ ( A ) ) , \boxed{
d_1(A)
=
d_2(\varphi(A)),
} d 1 ( A ) = d 2 ( φ ( A )) ,
m 1 ( B ) = m 2 ( φ ( B ) ) , \boxed{
m_1(B)
=
m_2(\varphi(B)),
} m 1 ( B ) = m 2 ( φ ( B )) ,
ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) . \boxed{
\phi_i^{(1)}
=
\phi_{\varphi(i)}^{(2)}.
} ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) .
所以 freedom algebra 的核心不是 labels 所有。
對 lossless representation:
ρ \rho ρ
亦有:
I ρ ( ρ ( ω ) ) = I ( ω ) . \boxed{
I_\rho(\rho(\omega))
=
I(\omega).
} I ρ ( ρ ( ω )) = I ( ω ) .
因此 radix、symbol names、incidence ordering 等都可以被識別為 representation layer,而不必被誤認成 underlying relational invariant。
baseline example 更顯示單一 quotient class 可對應:
15400 15400 15400
個不同 labeled hypergraphs。
所以本文一句話總結:
The representation may change; the generator quotient need not. \boxed{
\textbf{The representation may change; the generator quotient need not.}
} The representation may change; the generator quotient need not.
中文:
表示可以改變;真正的 generator quotient 不必因此改變。
Appendix A. 核心定義
Hypergraph isomorphism:
φ : H 1 ≅ H 2 . \varphi:
\mathcal H_1
\cong
\mathcal H_2. φ : H 1 ≅ H 2 .
Generator equivalence:
X 1 ∼ q X 2 . X_1\sim_q X_2. X 1 ∼ q X 2 .
Generator Quotient:
Q q ( X ) = [ H q ( X ) ] ≅ . \mathfrak Q_q(X)
=
[
\mathcal H_q(X)
]_{\cong}. Q q ( X ) = [ H q ( X ) ] ≅ .
State equivariance:
V 1 ( A ) = V 2 ( φ ( A ) ) . V_1(A)
=
V_2(\varphi(A)). V 1 ( A ) = V 2 ( φ ( A )) .
Latent equivariance:
d 1 ( A ) = d 2 ( φ ( A ) ) . d_1(A)
=
d_2(\varphi(A)). d 1 ( A ) = d 2 ( φ ( A )) .
Möbius equivariance:
m 1 ( B ) = m 2 ( φ ( B ) ) . m_1(B)
=
m_2(\varphi(B)). m 1 ( B ) = m 2 ( φ ( B )) .
Shapley equivariance:
ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) . \phi_i^{(1)}
=
\phi_{\varphi(i)}^{(2)}. ϕ i ( 1 ) = ϕ φ ( i ) ( 2 ) .
Appendix B. Formal Status
解析定理:
Theorem 4.1;
Theorem 6.1;
Theorem 7.1;
Theorem 8.1;
Theorem 9.1;
Theorem 11.1;
Theorem 16.1;
Theorem 19.1;
Corollaries 16.2、20.1;
Proposition 23.1。
Exact computational audit:
E5 全部 4096 4096 4096 subset states 的 state / latent / Möbius equivariance;
Shapley equivariance;
radix 2 , 3 , 8 , 10 , 16 , 36 2,3,8,10,16,36 2 , 3 , 8 , 10 , 16 , 36 encode/decode audit;
15400 15400 15400 labeled orbit enumeration;
E4 straight / zigzag separated equivalence;
coupled negative control。
尚未宣稱:
finite Generator Quotient 是 universal spatial invariant;
aggregate fingerprints 是 complete invariants;
lossy coarse-graining 保持 quotient;
different-resolution hypergraphs 普通同構;
physical ontology identity。