Enumerating Posterior-Supported Priors: First-Order Error of Finite-Integral Cross-Categorical Measurement Morphisms
Organizations: Shanghai Jiao Tong University · Shanghai Advanced Research Institute, Chinese Academy of Sciences
Abstract
In generative modeling, standard measure theory adopts the a.e. convention, which legitimately ignores any error component carried by a null-support set: a single null element can hide out-of-domain error inside null support and never be explicitly accounted for. The contribution of this paper is not a new distributional distance, but rather a verifiable structure that writes down which underlying predicate supports a high-level object, when it is invertible, and at which layer the error resides, from which we propose the enumerating posterior-supported prior framework. The framework operates on a three-layer categorical chain Quot->MorTop->Set: the first two layers posit no point elements and take binary-predicate quotient partitions as objects; the replica-index counting measure successively discriminates one-to-many collections into injections, lifting object collection from surjection to a quotient-layer bijection; the mutual exclusion of preimages in the shared measure space is resolved by an invertible MPT. Four results consolidate into Thm. 7: binary quotient-partition meta-categorical objects support the set categorical functional along categorical support via measure descent--recovery, enumerating the morphism itself rather than its preimage and object; their multiplicity is normalized through replica-index injection and accounted as an explicitly first-order error. The contribution of this paper is a transformation from nontrivial-input construction -> engineering-computable first-order output: the binary quotient-partition meta-categorical objects, supported via categorical support measure descent--recovery, constitute a substantive nontrivial input---precisely because they lift the out-of-domain error that hid inside null support into a first-order falsifiable coefficient mu_err, a finitely obtainable finite integral first-order error.
Figures & tables
| Route | Invertible / pairing | Error tracking | Open problem solved here |
| Normalizing flow flow matching | point-wise invertible, rep.-dependent | Jacobian, no replica tracking | null-embedding ambiguity |
| Optimal transport metric comparison | coupling, not predicate bijection | only volume diff. | training–generation pairing bias |
| Diffusion model | not aimed at invertible pairing | no first-order error layering | iterative high-order error accumulation |
| Ours | quotient-layer bijection + invertible MPT | explicit first-order accounting, 3 channels close | all three |
| Layer | Object | Point elem. | Measure source | Failure if removed |
| binary predicate / quotient partition | none | none | Null-embedding ambiguity recurs | |
| quotient block / differential quotient | none | induced by partition or bijection | First-order error degenerates to high order | |
| derived instance | yes | pre-given | No accountable reading |
| Layer | Predicate form | Mutual-exclusion form | Compositional restriction |
| partition category | quotient-predicate partition (unique categorical object) | finite preimage mutual-exclusion constraint (see Corollary 10a) | no point elements; Boolean decision only |
| mapping category | infinite predicate | layered mutually-exclusive additive group | quotient-class bijection; measure via descent |
| set category | numerical functional | measure-space topological mutual exclusion | derived instances; dual measure |
| Chain | Content | Basis | Assumption attribution |
| ① | Assumption 1 output Postulates 1–4 | §2.4.3 | unique mathematical assumption |
| ② | Postulate 1, Postulate 2 Lemma 1(i) Definition 24 Principle 1, Principle 2 | this entry, Principle 1, Principle 2 | via Assumption 1 |
| ③ | Principle 3 Theorem 1 Axiom 2 predicate-matching evaluation bijection; point-/differential-reading divide | Theorem 6, Definition 19a, Premise 1 domain | Conventions 1–4 architectural; differential reading needs no Assumption 1 |
| ④ | Principle 4 Theorem 6(i) Condition 1 Lemma 6a(3) existence of a gradient direction decreasing first-order error (monotonic non-increase not guaranteed) | Lemma 6a, B.109 | Condition 1 is an implementation-layer per-round condition (registered in B.4), not a mathematical assumption; monotonicity claim abandoned |
| ⑤ | Principle 3, Principle 4 Proposition 20 B.113 cardinality reading; three-point distinction (see Definition 25c) | §6.3.2, Definition 25c | midpoint selection is definitional selection (see Definition 25c; three points: boundary midpoint / subtracted saddle , integral boundary , gradient direction ) |
| ⑥ | Principle 2 Proposition 21b Gaussian cylindrical-kernel compatible-family boundary (trace-class) | §5.3 | via Postulate 4 |
| Direction | Geometric key | Categorical motion | Measure motion | Arrival layer |
| To element (up) | Euclidean raise | lifting, lift-equip ( ) | lifting recovery ( , read numbers/points) | point element |
| To partition (down) | Euclidean lower | categorical descent ( , canonical descent functor) | measure descent (pushforward) | quotient block |
| Reading | Level | Counted object | Share caliber | Error |
| (i.e. ) | , morphism not equipped with global counting measure | 2 differential elements, each counted | block share (two blocks each count ) | none |
| , morphism equipped with global counting measure and lift-equipped via categorical support | union subset base 3 , , ; binomial hit base 2 (midpoint crossed both sides hit twice each , summing to one) | cardinality reading (hit base / subset base) | none (counting measure no error) | |
| Volume ratio | Lebesgue geometric measure | geometric volume | volume share | carries first-order error |
| Factor / mechanism of ( 26 ) | Code object | Implementation-layer reading | Basis |
| ①outer-layer Lebesgue marginal measure (after square root) | outer-layer Lebesgue tensor (see ( 24 )) | B.116(a), B.116(d) | |
| ②middle-layer truncated-Gaussian area | ( engineering reading ) | middle-layer truncated-Gaussian area (see ( 24 )) | B.116(b), B.116(d) |
| ③equivalence-class sign and replica-index-counting measure | normalized counting-measure reading on the equivalence class (zero error) | Definition 17, §6.3.1 | |
| (not entering theoretical ( 26 )) | sign-bit mask , MSE lower bound | implementation-layer gating / training-period lower-bound proxy | implementation layer |
| Invoking conclusion | Invoked conclusion | Direct premise dependency | Cyclic |
| Lemma 6a(3) (see §6.4.1) | Theorem 6(i), Theorem 6(e)–(f) | Conventions 1–4, Definition 19a, Lemma 1(i); point-level readings additionally carry Premise 1 | No |
| Lemma 6a(3) (see §6.4.1) | Proposition 21, Proposition 21b (cylindrical-kernel-family reading) | Postulate 4 via Assumption 1 | No |
| Lemma 6a(3) (see §6.4.1) | Measure-comparison Condition 1 | implementation-layer round condition, not a mathematical premise of this paper | No (implementation condition attributed separately) |
| Definition 25c (see §6.4.2) | Theorem 6(a)(d)(i), Definition 19a | Conventions 1–4; differential reading does not carry Premise 1 | No |
| B.109 | Theorem 6, Proposition 21 | same two items as above | No |
| B.111 | Definition 5, Definition 24, Lemma 1, Definition 25c | Definition 5 (partition-induced), Definition 24 (pushout–pullback), Lemma 1(i) (block-bijection-group transitivity), Definition 25c (definitional selection) | No |
| Route | Main reading | Invertibility/ pairing | Structural difference | Our improvement |
| Normalizing flow | Invertible transport and Jacobian | Point-wise invertible; sample pairing still depends on representation | Does not handle surjection-cover completeness of the element-free quotient layer | Null-embedding ambiguity : surjective morphisms, after being successively discriminated into injections by the replica-index measure, are lifted to a quotient-layer bijection; predicate matching gives a bijection criterion [ 37 ; 8 ] [*1] |
| Optimal transport / metric measure comparison | Displacement, kinetic energy, or structural distance | Usually yields couplings rather than predicate bijections | Does not track replicas via quotient-block predicate words | Training–generation pairing bias : instance-level pairing error becomes a measurable, accountable object [ 43 ; 4 ; 31 ] [*2] |
| Diffusion model | Forward kernel and score | Generally not aimed at invertible pairing | Does not directly give first-order error layering | Iterative high-order error accumulation : first-order error is explicitly localized; high-order channels close via three-channel verification (see Proposition 23c) [ 3 ; 33 ; 39 ; 40 ; 16 ] [*3] |
| Ours | Quotient-partition predicates and dual-measure descent–recovery | Quotient-layer bijection and invertible MPT | Uses categorical support instead of point-wise representation to organize the three-layer relations | The three problems obtain a structural resolution at the same reading layer, and the sole backbone mathematical assumption is Assumption 1 (implementation-layer conditions are registered separately and do not enter the backbone assumption list) |
| Item | Standard result (citation/basis) | Our new treatment / new proposition |
| Theorem 1 (quotient equivalence class element-disjoint no collapse functional well-defined) | Main statement: pairwise disjoint quotient-block family (no collapse) equivalence-class differential definable as a functional; the finite equal-cardinality special case reduces to “no collapse quotient-set bijection block-pairing bijection” | Block-layer premise for Theorem 7 / Corollary 13; quotient-bijection criterion (modulo label permutation) |
| Lemma 1(i) (block counting measure unique) | Unique invariant probability measure on a finite transitive group (Haar special case) | As a “partition-induced reading” rather than an externally imposed reference measure (B.7.6) |
| Lemma 4 (differential–predicate bijection tower) | Binary tree–binary word one-to-one correspondence (combinatorial) | As an inter-layer lifting channel, routing differentials into quotient blocks via predicate words |
| Proposition 21a (Gaussian kernel uniqueness) | Schoenberg / Bochner / Maxwell type standard theorem | Adaptation of uniqueness conditions (iii-1)(iii-2) + treatment of singular positive-semidefinite degeneracy |
| Proposition 21b ( -additive trace-class) | Minlos / Gross standard result | As applicability-domain boundary (finite-dimensional vs cylindrical-measure compatibility family) |
| Replica-index measure injection + first-order normalization | Replica counting + pushforward many-to-one irreversible (standard measure theory) | New to this paper : one-to-many collection injection; multiplicity normalized into first-order error |
| Theoretical claim | Experimental reading | What it can support | What it cannot alone derive |
| Quotient-block partition compatible with semantic categories | Two MNIST 1NN columns and latent dimension | Empirical compatibility of clustering structure under training/test 1NN | Independent proof of quotient bijection or inner compactness |
| Matching occurs at the measure-structure level rather than training-set replication | Joint 1NN and RMSE reading, with self-mapping cancelled | Clues of structural matching without value-by-value replication | Proving higher diversity solely from larger RMSE |
| Zero-residue anchor correlated with pairing residual | CIFAR-10 learned saddle-point gap | Empirical evidence of gap improvement under the same structure | Uniqueness, SGD reachability, or convergence rate |
| Deviation between mid-layer zero-residue anchor and outer variational Gaussian vertex is monitorable | Figure 4 Lebesgue excess curve and finite-loss reading | Finiteness, computability, and training-process diagnosis of the generative-side mid-outer-layer excess | Monotone convergence, uniqueness, SGD reachability, or high-order error closure at first order |
| Model | 1NN | Error rate ( ) | RMSE | Latent dim |
| Posterior-enumerated prior | 98.17% 0.32% / 97.51% 0.39% (training-set 1NN / test-set 1NN) | 1.83% 0.32% / 2.49% 0.39% | 0.172 | 50 |
| ALAE [ 34 ] | / 94.59%(50) / 97.47%(1024) | / 5.41%(50) / 2.53%(1024) | 50 / 1024 | |
| Cold Diffusion [ 3 ] | 0.154 | 784 (pixel space) | ||
| BiGAN [ 9 ] | 96.81% / 97.39% | 3.19% / 2.61% | 200 | |
| AE [ 9 ] | 97.63% / | 2.37% / | 200 | |
| TreeVAE [ 30 ] | (clustering ACC 90.2%) / | 8 |
| Model | Learned saddle-point gap | Converged saddle-point gap |
| Posterior-enumerated prior | 0.106 0.81% | 0 |
| DDPM [ 16 ] | 0.163 | 0 |