Finite filtration: checking information nesting by partition refinement, and separating it from adaptedness and independence

On this page
  1. Goal and scope
  2. Givens and symbols
  3. Rules and sources
  4. Main derivation
  5. Substeps and lemmas
  6. Reverse audits
  7. Error controls and negative controls
  8. Result and limitations
  9. Verification and source binding

← Guide index · Original MKO

derivation-mko-finite-filtration-01 · production-20261019-adopted-96

Goal and scope

On Ω={0,1,2,3} (two coin tosses: 0=HH, 1=HT, 2=TH, 3=TT), represent the σ-algebras at times 0, 1, 2 by partition rows P_0, P_1, P_2. Prove that 'the later partition refines the earlier one ⇔ the σ-algebras are nested', check every adjacent pair of times completely and conclude that (F_t) is a filtration; give a sequence of partitions that is not a filtration; and show that a filtration neither makes every process adapted nor implies independence.

Works inside the MKO's scope: every partition uses the same outcomes 0 to 3, and labels within a row indicate only block equivalence (assumption a1); refinement is checked completely at every pair of adjacent times (assumption a2). No natural-filtration, independence or Markov claim is made; martingales are not treated; the canonical MKO and its English edition are not changed.

Givens and symbols

  • set_operations_def
  • subset_def
  • set_extensionality
  • prob_axioms
  • G1 — Ω={0,1,2,3}; meaning of the outcomes: 0=HH, 1=HT, 2=TH, 3=TT (for orientation only; it does not affect the partitions).
  • G2 — Partition rows (labels for outcomes 0, 1, 2, 3): P_0=(a,a,a,a), P_1=(a,a,b,b), P_2=(a,b,c,d); equal labels mean the same block (MKO assumption a1).
  • G3 — Definitions: F_t=σ(P_t); (F_t)_{t=0,1,2} is a filtration if s≤t ⇒ F_s⊆F_t. P' refines P (written P'⪯P) if every block of P' is contained in some block of P.
  • G4 — Contrast sequence: Q_0=(a,a,b,b), Q_1=(a,b,a,b). Processes: X_1(ω)=1 if ω∈{0,1} (first toss heads), else 0; Y_1(ω)=1 if ω∈{0,2} (second toss heads), else 0.
  • P_t — the partition at time t
  • F_t=\sigma(P_t) — the σ-algebra generated by P_t (unions of blocks)
  • P'\preceq P — P' refines P: each block of P' lies inside some block of P
  • X_1,\ Y_1 — indicators that the first, respectively second, toss is heads
  • Q_0,Q_1 — a sequence of partitions that is not a filtration

Rules and sources

  • set_operations_def · prerequisite — Set operations: x∈A∪B ⇔ x∈A or x∈B; x∈A∩B ⇔ x∈A and x∈B; x∈A∖B ⇔ x∈A and x∉B; in a universe U, Aᶜ=U∖A.
  • subset_def · prerequisite — Subset: A⊆B means every element of A is an element of B, i.e. ∀x (x∈A⇒x∈B).
  • set_extensionality · prerequisite — Set equality: A=B if and only if A⊆B and B⊆A, i.e. they have the same elements.
  • prob_axioms · prerequisite — Probability axioms: P(A)≥0; P(Ω)=1; the probability of a (countable) union of pairwise disjoint events is the sum of their probabilities. Hence P(∅)=0 and P(Aᶜ)=1−P(A).
  • set_builder · prerequisite — Set-builder notation: x∈{y∈A ∣ P(y)} if and only if x∈A and P(x).
  • def_unfold · prerequisite — Unfolding a definition: replacing a defined term by its definition (or folding it back) does not change whether a statement is true.
  • numeric_evaluation · prerequisite — Exact evaluation of an explicit integer or rational expression; independently recomputable.
  • forall_intro · prerequisite — ∀-introduction (generalization): if a statement is proved for an arbitrary element x, with no extra assumption on x, it holds for every x in the domain.
  • logic_chain · prerequisite — Chaining: from P⇒Q and Q⇒R infer P⇒R; likewise for ⇔.
  • contradiction · prerequisite — Proof by contradiction: if assuming ¬P leads to a contradiction (a statement and its negation both hold), then P holds.
  • quantifier_negation · prerequisite — Negating quantifiers: ¬(∀x, P(x)) is equivalent to ∃x, ¬P(x), and ¬(∃x, P(x)) is equivalent to ∀x, ¬P(x).
  • and_intro_elim · prerequisite — Conjunction: from P and Q infer P∧Q; from P∧Q infer P and infer Q.
  • mko_filtration_def · cited — Represent each time's sigma-algebra by a partition of the same finite sample space. F_s being contained in F_t means the later partition refines the earlier one. A filtration specifies information nesting only: a process is not automatically adapted, and no natural-filtration, independence or Markov claim follows. · mko-finite-filtration
  • sigma_partition · prerequisite — The σ-algebra σ(𝒫) generated by a partition 𝒫 of a finite set Ω consists exactly of the unions of blocks (including ∅); it has 2^{|𝒫|} events.
  • meas_const · prerequisite — A real function on Ω is σ(𝒫)-measurable exactly when it is constant on each block of 𝒫.
  • L_refine_iff · lemma — For partitions P and P' of the same finite Ω: P'⪯P ⇔ σ(P)⊆σ(P').

Main derivation

S01 · equivalent

Before

P_0=(a,a,a,a),\ P_1=(a,a,b,b),\ P_2=(a,b,c,d)

After

P_0=\{\Omega\},\ P_1=\{\{0,1\},\{2,3\}\},\ P_2=\{\{0\},\{1\},\{2\},\{3\}\}

Conditions: none

Rule: set_builder · Uses: G2, set_builder

Leaf basis: set_builder + def_unfold

The block of ω is {ω'∈Ω ∣ ω' has the same label as ω}. In P_0 all labels agree, giving the single block Ω; in P_1, 0 and 1 share a and 2 and 3 share b; in P_2 the four labels differ, giving four singletons. The label names carry no meaning: writing P_1 as (b,b,a,a) gives the same partition.

S02 · equivalent

Before

F_t=\sigma(P_t)

After

F_0=\{\varnothing,\Omega\},\ F_1=\{\varnothing,\{0,1\},\{2,3\},\Omega\},\ F_2=2^{\Omega}

Conditions: none

Rule: sigma_partition · Uses: S01, sigma_partition

Leaf basis: sigma_partition + set_operations_def + numeric_evaluation

A σ-algebra consists of the unions of blocks: F_0 has 2^1=2 events, F_1 has 2^2=4, and F_2 has 2^4=16, all subsets of Ω (N1 to N3). An event A is measurable at time t exactly when it is a union of blocks of P_t (MKO paragraph 2).

S03 · equivalent

Before

P'\preceq P

After

\sigma(P)\subseteq\sigma(P')

Conditions: P,P'\ \text{partitions of the same}\ \Omega

Rule: L_refine_iff · Uses: G3, sigma_partition

Leaf basis: sigma_partition + set_extensionality + subset_def

By Lemma L_refine_iff (proved in S03.1 and S03.2), refinement is equivalent to inclusion of σ-algebras. This recasts inclusion between finite event families as a more compact complete check of block inclusions. Here |Ω|=4, so there are at most 2^4=16 events; a partition with b blocks generates exactly 2^b union events, not infinitely many events.

S04 · implies

Before

P_1\ \text{vs}\ P_0

After

\{0,1\}\subseteq\Omega,\ \{2,3\}\subseteq\Omega:\ P_1\preceq P_0

Conditions: none

Rule: subset_def · Uses: S01, G3

Leaf basis: subset_def + forall_intro

Check both blocks of P_1: {0,1} and {2,3} each lie inside Ω, the only block of P_0. The check is complete; every block was examined.

S05 · implies

Before

P_2\ \text{vs}\ P_1

After

\{0\},\{1\}\subseteq\{0,1\},\ \{2\},\{3\}\subseteq\{2,3\}:\ P_2\preceq P_1

Conditions: none

Rule: subset_def · Uses: S01, G3

Leaf basis: subset_def + forall_intro

Check all 4 blocks of P_2: {0} and {1} lie in {0,1}, and {2} and {3} lie in {2,3}. The check is complete.

S06 · implies

Before

P_1\preceq P_0,\ P_2\preceq P_1

After

F_0\subseteq F_1\subseteq F_2,\ \text{so}\ (F_t)_{t=0,1,2}\ \text{is a filtration}

Conditions: none

Rule: L_refine_iff · Uses: S03, S04, S05

Leaf basis: subset_def + logic_chain + mko_filtration_def

By L_refine_iff, S04 gives F_0⊆F_1 and S05 gives F_1⊆F_2; inclusion is transitive, so F_0⊆F_2. Hence F_s⊆F_t for all s≤t (trivially when s=t), and (F_t) is a filtration. Checking adjacent times suffices precisely because inclusion is transitive (MKO assumption a2).

S07 · implies

Before

A=\{0,1,2\}

After

A\in F_2,\ A\notin F_1

Conditions: none

Rule: sigma_partition · Uses: S02

Leaf basis: sigma_partition + contradiction

A is a union of singletons, so A∈F_2. If A were a union of P_1-blocks, then since 2∈A and the block of 2 is {2,3}, A would contain 3; but 3∉A, so A∉F_1. Information grows with time: an event decidable at time 2 need not be decidable at time 1.

S08 · implies

Before

Q_0=\{\{0,1\},\{2,3\}\},\ Q_1=\{\{0,2\},\{1,3\}\}

After

Q_1\not\preceq Q_0,\ \text{so}\ \sigma(Q_0)\not\subseteq\sigma(Q_1)

Conditions: none

Rule: L_refine_iff · Uses: G4, S03, set_builder

Leaf basis: subset_def + quantifier_negation + sigma_partition

The Q_1-block {0,2} lies neither in {0,1} (2∉{0,1}) nor in {2,3} (0∉{2,3}), so Q_1 does not refine Q_0; by L_refine_iff, σ(Q_0)⊄σ(Q_1). A concrete witness: {0,1}∈σ(Q_0), but it contains 0 and not 2, while {0,2} is a Q_1-block, so {0,1} is not a union of Q_1-blocks. The information at the two times is not nested: an arbitrary sequence of partitions is not a filtration.

S09 · implies

Before

X_1=(1,1,0,0)\ \text{on}\ (0,1,2,3)

After

X_1\ F_1\text{-measurable}

Conditions: none

Rule: meas_const · Uses: G4, S01, meas_const

Leaf basis: meas_const

X_1 equals 1 throughout {0,1} and 0 throughout {2,3}, i.e. it is constant on every block of P_1, so X_1 is F_1-measurable: the first toss is known at time 1.

S10 · implies

Before

Y_1=(1,0,1,0)\ \text{on}\ (0,1,2,3)

After

Y_1\ \text{not}\ F_1\text{-measurable}

Conditions: none

Rule: meas_const · Uses: G4, S01, S06, meas_const

Leaf basis: meas_const + quantifier_negation

Y_1(0)=1 and Y_1(1)=0, while 0 and 1 lie in the same block of P_1, so Y_1 is not constant there and is not F_1-measurable. Hence a process whose time-1 value is Y_1 is not adapted to (F_t), even though S06 proved that (F_t) is a filtration. Y_1 is F_2-measurable (every function is constant on singletons).

S11 · implies

Before

P(\{0\})=P(\{3\})=\tfrac12,\ P(\{1\})=P(\{2\})=0

After

P(X_1=1,\,Y_1=1)=\tfrac12\neq\tfrac14=P(X_1=1)P(Y_1=1)

Conditions: none

Rule: prob_axioms · Uses: S06, G4, prob_axioms

Leaf basis: prob_axioms + numeric_evaluation + quantifier_negation

The definition of (F_t) uses no probability at all. Take a valid probability (total 1/2+0+0+1/2=1, N5): P(X_1=1)=P({0,1})=1/2 and P(Y_1=1)=P({0,2})=1/2, but P(X_1=1,Y_1=1)=P({0})=1/2≠1/4 (N4). The two tosses are not independent, while the filtration is unchanged. So a filtration does not imply independence; likewise it implies no Markov property.

S12 · implies

Before

\text{S06, S08, S10, S11}

After

(F_t)\ \text{filtration};\ (\sigma(Q_t))\ \text{not};\ \text{adaptedness and independence not implied}

Conditions: none

Rule: mko_filtration_def · Uses: S06, S08, S10, S11

Leaf basis: and_intro_elim

The complete adjacent refinement check proves that (F_t) is a filtration; the sequence Q is not, because one block fails to refine; a filtration expresses information nesting only, a process with Y_1 as its time-1 value is not adapted, and the tosses need not be independent.

Substeps and lemmas

S03.1 · implies

Before

P'\preceq P,\ B\in P

After

B=\bigcup\{C\in P'\mid C\cap B\neq\varnothing\},\ \text{hence}\ \sigma(P)\subseteq\sigma(P')

Conditions: none

Rule: set_extensionality · Uses: G3, sigma_partition, set_operations_def

Leaf basis: set_extensionality + set_operations_def + sigma_partition

(⇒) Take B∈P. ⊆: each ω∈B lies in some block C of P', and C∩B contains ω. ⊇: if C∩B≠∅, refinement gives C⊆B' for some B'∈P; then B'∩B⊇C∩B≠∅, and blocks of P are equal or disjoint, so B'=B and C⊆B. Hence each block of P is a union of blocks of P'; unions of P-blocks are then unions of P'-blocks, so σ(P)⊆σ(P').

S03.2 · implies

Before

\sigma(P)\subseteq\sigma(P'),\ C\in P',\ \omega\in C

After

C\subseteq B\ \text{for the block}\ B\in P\ \text{containing}\ \omega

Conditions: none

Rule: subset_def · Uses: G3, sigma_partition

Leaf basis: subset_def + sigma_partition + set_operations_def

(⇐) Let B be the block of P containing ω. B∈σ(P)⊆σ(P'), so B is a union of blocks of P'; ω∈B lies in one of them, and ω belongs only to the P'-block C, so C is one of them and C⊆B. This holds for every C, so P'⪯P. S03.1 and S03.2 together prove Lemma L_refine_iff.

Reverse audits

  • S12 · sufficient_goal_reduction · S06, S08, S10, S11 — Goal G: the four conclusions stated in the goal. Sufficient condition H: S06, S08, S10 and S11; H⇒G by conjunction. H is proved forwards from G1 to G4; G is never used as a premise.
  • S11 · premise_audit · S06, prob_axioms — One counterexample suffices to refute 'a filtration implies independence'; it is not claimed that independence fails under every probability.
  • S10 · premise_audit · meas_const, S06 — Adaptedness is a separate condition linking a process to a filtration and must be checked on its own (MKO paragraph 2); the existence of a filtration does not supply it.
  • S09 · premise_audit · meas_const, S01 — Constancy must be checked block by block; here both blocks pass.
  • S08 · premise_audit · G4, L_refine_iff — One failing block suffices to refute refinement; the witness event independently confirms that inclusion fails.
  • S07 · premise_audit · S02, sigma_partition — One split block, {2,3}, suffices; no converse is claimed.
  • S06 · premise_audit · S03, S04, S05, subset_def — Every adjacent pair of times must pass; it is not claimed that this filtration is the natural filtration of any process.
  • S05 · premise_audit · S01, subset_def — All 4 blocks pass; a single failure would rule out refinement.
  • S04 · premise_audit · S01, subset_def — Refinement requires every block of P_1 to pass; here both do.
  • S03.2 · premise_audit · G3, sigma_partition, subset_def — The key is that the blocks of P' are disjoint, so the P'-block containing ω is unique.
  • S03.1 · premise_audit · G3, set_extensionality, sigma_partition — It uses that P' covers Ω and that the blocks of P are disjoint; only one direction is proved here.
  • S03 · inverse_equivalence · L_refine_iff — The two directions are proved in S03.1 and S03.2; both partitions must be on the same Ω.
  • S02 · inverse_equivalence · sigma_partition — The listed sets are exactly all unions of blocks; the step holds both ways.
  • S01 · inverse_equivalence · set_builder, G2 — From the blocks a label row can be rebuilt (one new label per block), unique up to renaming labels; both descriptions determine the same partition.

Error controls and negative controls

An arbitrary partition sequence need not be a filtration

The MKO's misconception: calling any sequence of partitions a filtration. Q_0 and Q_1 are both partitions of the same Ω, but Q_1 does not refine Q_0, and {0,1} is decidable at time 0 but not at time 1 (S08). Every block must be checked at every adjacent pair of times.

A filtration implies no adaptedness, independence or Markov property

The MKO's second misconception. S06 proves that (F_t) is a filtration, yet Y_1 is not F_1-measurable (S10), and under P({0})=P({3})=1/2 the tosses are not independent (S11). Adaptedness must be checked separately for each process; independence and Markov properties depend on the probability, which the definition of a filtration does not involve.

Sampled events do not prove inclusion

MKO assumption a2: finitely sampled events do not establish inclusion of σ-algebras. For the sequence Q, the events ∅ and Ω lie in both σ(Q_0) and σ(Q_1); sampling only these would wrongly suggest inclusion. The complete block-refinement check (the lemma of S03) is the certificate.

  • σ(Q_0) and σ(Q_1) are σ-algebras on the same Ω, so (σ(Q_0),σ(Q_1)) is a filtration → reject — {0,1}∈σ(Q_0) but it is not in σ(Q_1) (S08).
  • (F_t) is a filtration, so the second toss Y_1 is F_1-measurable → reject — Y_1 takes the values 1 and 0 on the block {0,1} (S10).
  • (F_t) is a filtration, so the two tosses are independent → reject — With P({0})=P({3})=1/2, P(both heads)=1/2≠1/4 (S11).
  • ∅ and Ω both belong to σ(Q_1), so σ(Q_0)⊆σ(Q_1) → reject — Inclusion must hold for every event of σ(Q_0); {0,1} fails.
  • Rewriting P_1 as (b,b,a,a) gives a different partition and a different F_1 → reject — Labels only indicate block equivalence; the blocks are still {0,1} and {2,3} (S01).

Result and limitations

For partitions of the same finite Ω, the later partition refines the earlier one exactly when the σ-algebras are nested. The 2 blocks of P_1 lie in Ω and the 4 singletons of P_2 lie in blocks of P_1, so F_0⊆F_1⊆F_2 and (F_t) is a filtration; the Q_1-block {0,2} lies in no Q_0-block, {0,1}∈σ(Q_0)∖σ(Q_1), and Q is not a filtration. The first toss X_1 is F_1-measurable, the second toss Y_1 is not; under P({0})=P({3})=1/2 the two tosses are not independent. A filtration expresses information nesting only.

  • This guide does not claim that (F_t) is the natural filtration of any process, and it does not treat martingales (the MKO's next step, to be added together with conditional expectation).

Verification and source binding

5 declared checks for this guide; only the listed expressions are covered.

  • source commit: b28a5edbd46279b1db2a3f3f48bb3f447e18899f
  • adopted source SHA-256: e6820be99eb83c1217b02247f1bc0e38f7ba1bd24ea85ba09fd8e6937818f6c0
  • original source SHA-256: ab8a8b045d8752cb59b7afd124424df8d7210731678c8341756298606f15afcf
  • Lean: NOT_RUN · independent AI: NOT_MEASURED

Same-source full JSON (formulas, ASTs, controls and revision log)