Invariant Sigma-Algebra¶

The sub-sigma-algebra of measurable events unchanged by a specified measurable transformation or action—exactly or modulo null sets—encoding all event-level information that the dynamics cannot alter.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2096
Origin domain: ergodic theory and probability
Subdomain: invariant events and ergodic decomposition
Aliases: Sigma-algebra of invariant events, Invariant sigma-field

Core Idea¶

Given a measurable space and a specified measurable transformation or action, the invariant sigma-algebra collects the measurable events whose truth is unchanged by that dynamics. For a measurable map \(T:X\to X\), the strict form is

\[ \mathcal I_T=\{A\in\mathcal F:T^{-1}A=A\}. \]

On a probability space, ergodic theory commonly works modulo null sets^[1]:

\[ \mathcal I_T^{\mu}=\{A\in\mathcal F:\mu(T^{-1}A\triangle A)=0\}. \]

Both collections are sigma-algebras. They contain the empty set and the whole space and are closed under complements and countable unions because inverse images commute with those operations. The object therefore does more than list isolated invariants: it organizes every measurable yes/no question that survives the action into one information structure.

The action must always be named. A set can be invariant under a time shift but not a coordinate permutation, or under one subgroup but not a larger group. The same underlying measurable space can consequently carry many invariant sigma-algebras. In measure-preserving dynamics, the system is ergodic exactly when its invariant sigma-algebra is trivial modulo the measure: every invariant event has probability zero or one.

Structural Signature¶

The abstraction contains eight roles:

the measurable space \((X,\mathcal F)\) — the state or sample space and its admissible events;
the dynamics or action — a measurable map \(T\), semigroup/monoid action, or group action whose inverse images act on events;
the event transform — \(A\mapsto T^{-1}A\), or the corresponding preimage under every action element;
the equality convention — strict set equality or equality modulo a declared null ideal;
the invariant events — events fixed under the event transform in the chosen sense;
the sigma closure — closure under complement and countable union, which makes the invariant collection a sub-sigma-algebra of \(\mathcal F\);
the retained information — random variables or factors measurable with respect to the invariant sigma-algebra, representing what is constant along the dynamics;
the triviality test — under a probability measure, whether every invariant event has probability zero or one, the criterion for ergodicity.

For an action \(\alpha_g:X\to X\) of a group or monoid \(G\), the strict version becomes

\[ \mathcal I_{\alpha}=\{A\in\mathcal F:\alpha_g^{-1}A=A\text{ for every }g\in G\}. \]

The invariant is: every member is measurable, fixed by every declared action element in the selected equality sense, and the entire collection remains closed under sigma-algebra operations.

What It Is Not¶

It is not an invariant measure. A measure is invariant when \(\mu(T^{-1}A)=\mu(A)\) for all measurable \(A\); individual events can move while retaining measure. The invariant sigma-algebra instead selects events whose membership is unchanged, exactly or almost surely.

It is not one invariant set. It is the full sigma-algebra of all invariant measurable events for the chosen action. A single invariant event generates a smaller sigma-algebra but need not determine the full object.

It is not the original sigma-algebra \(\mathcal F\) unless every measurable event is invariant. It is not automatically the trivial sigma-algebra either; triviality modulo \(\mu\) is the additional ergodicity property.

It is not the tail sigma-algebra. On a one-sided sequence space, every strictly shift-invariant event is a tail event, but a tail event need not be fixed by the shift. Tail measurability means insensitivity to finitely many initial coordinates, not necessarily invariance under reindexing the entire tail.

It is not the exchangeable sigma-algebra unless the declared action is the group of finite coordinate permutations. Shift and permutation actions yield different invariance questions.

Scope of Application¶

The home domain is ergodic theory, probability, and stationary stochastic processes. The construction applies to measure-preserving transformations, measurable group actions, shifts on path or sequence spaces, coordinate permutations, and related kernels under appropriately stated definitions.

In the strict measurable-space version no probability measure is required. Once a measure is present, null-set completion and almost-sure equality become central because random variables and conditional expectations are identified modulo null sets. Authors may use the same notation \(\mathcal I\) for strict events, almost-invariant events, or the completed sigma-algebra; a reference-grade account must state the convention.

The object is the conditioning sigma-algebra in ergodic theorems. In the pointwise ergodic theorem, time averages converge under standard hypotheses to conditional expectation given invariant information^[2]. In an ergodic system that conditional expectation is almost surely constant, reducing to the space average.

Permutation-invariant event sigma-algebras also organize zero-one laws and exchangeability. The action and probability assumptions determine which theorem applies; “invariant” alone licenses none of them.

Clarity¶

A recognition test asks:

What is the measurable space and ambient sigma-algebra?
Which transformation, semigroup, monoid, or group acts?
Does invariance mean strict equality, equality modulo null sets, or equality in the measure algebra?
Must an event be fixed by one generator or by every action element?
Is the measure preserved, nonsingular, or merely present?
What invariant random variables or factor does the sub-sigma-algebra encode?
Is a claimed triviality exact or only modulo probability-zero events?

For a finite cyclic permutation of four states with the full power set, invariant events are unions of whole cycles. If the action is one four-cycle, only \(\varnothing\) and \(X\) are invariant. If the permutation has two cycles, all unions of those two orbits form a four-element invariant sigma-algebra.

This example exposes the orbit logic without measure theory: event membership must be constant along every orbit. In a measured system, equality can be relaxed on null subsets of orbits.

Manages Complexity¶

A dynamical system generates indefinitely many time-indexed observations. The invariant sigma-algebra compresses that evolution into the information no amount of time shifting changes. Rather than analyze each invariant function separately, one conditions on a single sub-sigma-algebra that contains all invariant event information.

This compression clarifies ergodic decomposition. Nontrivial invariant events separate components that dynamics cannot mix across; conditioning on the invariant sigma-algebra records which component a state occupies. Within an ergodic component, invariant information is trivial.

It also turns qualitative preservation into standard probabilistic operations. Conditional expectation, measurability, zero-one laws, and factor maps can be applied to the invariant information structure. The closure property ensures that countable combinations of invariant questions remain legitimate events.

Diagnostic errors become local: using an invariant measure where an invariant event is required, switching strict and almost-sure equality, changing the action, or confusing tail insensitivity with shift invariance each changes a specific role.

Abstract Reasoning¶

If the acting family grows, the invariant sigma-algebra can only shrink: an event fixed by more transformations must satisfy more conditions. If the acting family is restricted to a subgroup, more events may become invariant. This order-reversing relation makes action scope load-bearing.

For two actions with invariant sigma-algebras \(\mathcal I_1\) and \(\mathcal I_2\), invariance under both lies in their intersection. Because intersections of sigma-algebras are sigma-algebras, common invariant information is again measurable structure.

Every \(\mathcal I_T\)-measurable real-valued function is constant along \(T\)-orbits in the appropriate sense, and invariant functions generate invariant events through inverse images of Borel sets. Modulo-null formulations require the usual care with representatives and completions.

Ergodicity predicts that invariant integrable random variables are almost surely constant. The converse is visible through indicators: if every invariant indicator is almost surely constant, every invariant event has probability zero or one.

On a product sequence space, changing finitely many coordinates preserves a permutation-invariant event under the finite-permutation action, while shifting deletes the first coordinate and relabels all others. The different actions explain why exchangeable, shift-invariant, and tail sigma-algebras must not be conflated.

Knowledge Transfer¶

Within ergodic theory, the same construction transfers from a single transformation to flows, \(\mathbb Z^d\)-actions, group actions, and stationary processes. The action changes; inverse-image fixed events and sigma closure remain.

In probability, finite-permutation invariance yields the symmetric or exchangeable event sigma-algebra used in the Hewitt–Savage zero-one law for independent identically distributed sequences^[3]. Shift invariance supplies the invariant information for stationary processes and time averages.

The conceptual core transfers to quotient and factor reasoning: invariant observables ignore motion within orbits and retain only distinctions between dynamically separated components. Yet the full node remains domain-specific because measurable spaces, sigma-algebras, null ideals, conditional expectation, and ergodicity are essential.

The portable primes are Invariance and Closure. Invariance supplies the fixed-under-action relation; Closure explains why complement and countable union preserve membership. Neither alone supplies the measurable-event information structure.

Examples¶

Finite orbit partition. A permutation acting on a finite set decomposes it into orbits. With the power-set sigma-algebra, invariant events are exactly unions of orbits. The resulting sigma-algebra records orbit identity and forgets position within an orbit.

Bernoulli shift. On an i.i.d. sequence space with the left shift and product measure, the shift is measure preserving and ergodic. The invariant sigma-algebra is trivial modulo null sets, so invariant events have probability zero or one.

Stationary but nonergodic mixture. First choose a latent bias, then generate a conditionally i.i.d. coin sequence. The shift cannot change the latent component. Invariant information can retain that component, and long-run averages converge to a component-dependent conditional expectation rather than one universal constant.

Finite-permutation action. The group of permutations moving only finitely many indices acts on an infinite sequence. Events fixed under all such permutations form the symmetric invariant sigma-algebra; under i.i.d. product measure, Hewitt–Savage makes it trivial modulo null sets^[3].

Non-example—invariant measure only. Rotation preserves Lebesgue measure on the circle, so every event keeps its measure under inverse image. Most arcs are moved to different arcs and are not invariant events.

Structural Tensions¶

Strict equality versus equality modulo null sets. Exact set invariance is cleaner as measurable structure; probability theory naturally identifies null modifications. The convention affects membership and must be explicit.

Fine information versus ergodic simplicity. A large invariant sigma-algebra describes many dynamically isolated components. Trivial invariant information enables strong time-to-space averaging but erases component distinctions.

One generator versus full action. For a group generated by transformations, invariance under generators may imply invariance under the group when inverse-image relations and measurability are handled correctly. For loosely specified families, silently changing the quantifier produces the wrong sigma-algebra.

Abstract factor versus concrete representatives. Modulo-null invariant events form clean equivalence classes in the measure algebra, while applications often need actual measurable representatives. Completion and version choices mediate the translation.

Related zero-one laws versus distinct symmetry classes. Tail, shift, and finite-permutation invariance can each lead to zero-one results under different assumptions. Similar conclusions do not make their sigma-algebras identical.

Structural–Framed Character¶

Invariant Sigma-Algebra is strongly structural–framed. It specifies the measurable domain, action, inverse-image operation, equality convention, fixed events, sigma closure, retained information, and triviality criterion. Each component is required to use the object in theorems or calculations.

Its identity is broader than any one theorem. Birkhoff's theorem, ergodic decomposition, stationary processes, and Hewitt–Savage use different actions and assumptions but reuse the same fixed-event information structure.

It is domain-specific rather than prime because sigma-algebras, null sets, conditional expectations, measurable functions, and ergodicity do not transfer literally outside measure theory and probability. The substrate-independent preservation core is already Invariance.

Structural Core vs. Domain Accent¶

The structural core is closed information retained under an action. A family of transformations operates; certain propositions remain fixed; closure combines them into an information algebra; and the resulting quotient ignores within-orbit change.

The domain accent adds measurable spaces, inverse images, complements and countable unions, probability-null equality, conditional expectation, invariant random variables, and ergodicity. These turn a generic collection of invariants into a sigma-algebra usable by probability theory.

Remove measurability and sigma closure and the residual is a family of invariant subsets. Remove the action and it is an arbitrary sub-sigma-algebra. Remove the equality convention and the object is ambiguous in measured settings. The combination supplies the autonomous domain identity.

Invariant Sigma-Algebra is a strict instance of Invariance. The named features are event-membership predicates; the named transformations act by inverse image; the preservation claim is strict equality or equality modulo null sets; and the inferential license is to reason through orbit-insensitive information.

It also instantiates Closure: invariant events contain the neutral events and remain closed under complement and countable union. Closure is essential to the sigma-algebra proof but does not identify which events are selected or which action is used.

The smallest prospective DAG placement is under Invariance. Closure remains a related-prime note rather than a second parent because the candidate's differentiating selection principle is preservation under dynamics.

Relationships to Other Abstractions¶

Current abstraction Invariant Sigma-Algebra Domain-specific

Parents (1) — more general patterns this builds on

Invariant Sigma-Algebra is a kind of Invariance Prime

Invariant Sigma-Algebra is a strict instance of Invariance.

Hierarchy path (1) — routes to 1 parentless root

Invariant Sigma-Algebra → Invariance

Neighborhood in Abstraction Space¶

Invariant Sigma-Algebra sits in a sparse region of the domain-specific corpus (80^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Invariant event or invariant set: one member of the invariant sigma-algebra.
Invariant measure: a measure preserved on all events, even events that move as sets.
Ambient sigma-algebra: all declared measurable events, whether invariant or not.
Trivial sigma-algebra: exactly \(\{\varnothing,X\}\); measure-triviality additionally allows null and conull events.
Tail sigma-algebra: events insensitive to every finite initial segment, not necessarily shift-fixed.
Exchangeable or symmetric sigma-algebra: invariant under finite coordinate permutations.
Shift-invariant sigma-algebra: the candidate specialized to a shift action.
Invariant function: a measurable function fixed under composition with the action; it is encoded by invariant-level events.
Ergodicity: the property that the invariant sigma-algebra is measure-trivial, not the sigma-algebra itself.
Invariant factor: the measurable factor represented by invariant information, often discussed modulo null sets.

References¶

[1] Walters, Peter. An Introduction to Ergodic Theory. Springer (Graduate Texts in Mathematics 79), 1982. Cited as a standard text on measure-preserving dynamics for the convention of working modulo null sets; the text was not reachable to confirm the wording, and the article's own Sheffield and Durrett references state the almost-invariance convention directly. registry ↩

[2] Durrett, Rick. Probability. Cambridge University Press, 2019. Durrett's Chapter 6 states Birkhoff's ergodic theorem (§6.2) after fixing the invariant sigma-field (§6.1), in the form the article uses: time averages converge almost surely and in L¹ to the conditional expectation given the invariant sigma-field. registry ↩

[3] Hewitt and Savage. “Symmetric measures on Cartesian products”. Transactions of the American Mathematical Society, 1955. Hewitt and Savage define symmetry by invariance under permutations 'leaving all but a finite number of integers fixed', and prove (Theorem 11.3) that a product measure on an infinite product takes only the values 0 and 1 on sets invariant under all finite coordinate permutations. The Hewitt–Savage zero-one law in its original form (Theorem 11.3): under a product measure on an infinite product, every event invariant under all finite permutations of the coordinates has probability zero or one. registry ↩a ↩b