Expectation Frames — @cite{veltman-1996} §4 #
@cite{veltman-1996}
Veltman calls §4 "the heart of the paper." Where §3 uses a single normality ordering for all domains, §4 introduces expectation frames: functions assigning a per-domain normality ordering. This enables conditional defaults — "if φ then normally ψ" — where the ordering is refined only for the φ-domain.
The key insight is the definition of normal worlds (Definition 4.3): a world w is normal in πd not just when w is optimal in d under πd, but when w is top-ranked in every subdomain d' ⊆ d containing w, under πd'. This is how specificity works — a more specific exception (governing a subdomain) can override a general default (governing a superdomain) without the general default needing to "know about" the exception.
Key definitions #
ExpFrame W: functionSet W → NormalityOrder WExpFrame.normal: Def 4.3 — check all subdomains, not just dExpFrame.coherent: Def 4.3(iii) — every non-empty d has normalsExpFrame.refineAt: Def 4.5 — refine pattern at domain d onlyFrameState: expectation state with a framecondDefault: Def 4.6 — conditional default update
Key results #
refineAt_preserves_coherent: Prop 4.7 — coherence characterizationrefineAt_comm_same: refinement at one domain commutesrefineAt_comm_diff: refinements at different domains commute
Connection to §3 #
The unconditional "normally ψ" is "(ψ ∨ ¬ψ) ⇝ ψ", so it refines the
frame at d = W. The ExpState from Basic.lean corresponds to a
FrameState with a constant frame. For constant frames with
connected orderings, the §4 normal definition reduces to §3's
optimal (since optimal and "top in all subdomains" coincide
when every ordering is connected).
An expectation frame: a function assigning a normality ordering to each domain (subset of worlds).
@cite{veltman-1996}, Definition 4.2: π assigns to every d ⊆ W an expectation pattern πd. Different domains may have different orderings — this is what enables conditional defaults.
- pattern : Set W → Core.Order.NormalityOrder W
The per-domain normality ordering
Instances For
The normal worlds in a domain under the frame.
@cite{veltman-1996}, Definition 4.3(i–ii): w is normal in πd iff w ∈ d and for every subdomain d' ⊆ d such that w ∈ d', w is at least as normal as every v ∈ d' under πd'.
This is stronger than just being optimal in d under πd: w must be top-ranked in every subdomain's pattern, not just the top-level one. This is how specificity works — a more specific exception (governing a subdomain) overrides a general default by making the world non-normal at the subdomain level.
Equations
Instances For
e is a default in πd iff d ∩ e ≠ ∅ and πd already respects e.
@cite{veltman-1996}, Definition 4.2(ii): e is a default in πd iff
d ∩ e ≠ ∅ and πd ∘ e = πd. Since refine_of_respects gives
the latter iff respects, we use respects directly.
Instances For
The constant frame: assigns the same ordering to every domain. This embeds §3's single-ordering setup into the §4 framework.
Equations
- Semantics.Dynamic.Default.ExpFrame.const no = { pattern := fun (x : Set W) => no }
Instances For
The total frame: assigns the total ordering to every domain. The initial state before any defaults have been processed.
Equations
Instances For
Frame refinement: refine the ordering at domain d only.
@cite{veltman-1996}, Definition 4.5(ii):
- π_{d∘e}(d') = πd' if d' ≠ d
- π_{d∘e}(d) = πd ∘ e
Unlike our earlier (incorrect) version that refined all d' ⊇ d,
this modifies only the pattern at d. The effect on superdomains
comes automatically through the normal computation, which checks
all subdomains — so a refinement at d changes which worlds count as
normal in any d' ⊇ d.
Equations
Instances For
An expectation state with frame: an information state paired with an expectation frame.
Veltman's §4 notation: σ = ⟨π, s⟩ where π is a frame and
s ⊆ W is the information state. This generalizes ExpState
from §3 (which uses a single ordering).
Instances For
Normal worlds in the current state: the most normal worlds in the info state according to the frame.
Instances For
The initial frame state: all worlds possible, total frame.
Equations
- Semantics.Dynamic.Default.FrameState.init = { info := Set.univ, frame := Semantics.Dynamic.Default.ExpFrame.total }
Instances For
Embed a §3 ExpState as a FrameState with constant frame.
Equations
- Semantics.Dynamic.Default.FrameState.ofExpState σ = { info := σ.info, frame := Semantics.Dynamic.Default.ExpFrame.const σ.order }
Instances For
Conditional default update: "if φ then normally ψ".
@cite{veltman-1996}, Definition 4.6. Given σ = ⟨π, s⟩:
- Let d = ‖φ‖ — the set of all φ-worlds (not just those in s)
- If d ∩ ‖ψ‖ = ∅, crash (the default is vacuous)
- If π_{d∘ψ} is incoherent, crash
- Otherwise, result is ⟨π_{d∘ψ}, s⟩
Note: d = ‖φ‖ over all of W, not ‖φ‖ ∩ s. The frame is about the space of possible worlds; the info state restricts which are epistemically accessible but doesn't determine the default domain.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Unconditional default: "normally ψ" = "(ψ ∨ ¬ψ) ⇝ ψ".
Refines the frame at d = W (all worlds).
Equations
- Semantics.Dynamic.Default.normallyFrame ψ σ = Semantics.Dynamic.Default.condDefault (fun (x : W) => True) ψ σ
Instances For
Assertion in the frame setting: eliminate non-φ-worlds.
Equations
Instances For
Presumably test in the frame setting: passes iff all normal worlds satisfy φ.
Equations
Instances For
Assertion preserves the frame.
Commutativity at one domain: refining by e₁ then e₂ at the same domain gives the same result as the reverse order.
Refinement can only remove normal worlds, never add them.
Because refineAt either preserves or strengthens the ordering
at each subdomain, the normal condition is harder to satisfy.
Coherence characterization (Proposition 4.7): given coherent π with d ∩ e ≠ ∅, the refined frame π_{d∘e} is coherent iff there is no d' ⊇ d such that nπd' ⊆ d \ e.
The condition "nπd' ⊆ d \ e" means all normal d'-worlds are in d but fail e. If this holds, refinement at d makes those worlds non-top (they don't satisfy e), potentially leaving no normals in the refined frame.
Note: nπd' here uses the original frame π's normals, not the refined frame's.
For a constant frame with a connected ordering, the §4 normal worlds in d coincide with §3's optimal worlds. This is because for connected orderings, "top in every subdomain" reduces to "optimal in d" — since optimal = top when the ordering is total or connected.
Applicability: a default e applies to s within frame π.
@cite{veltman-1996}, Definition 4.9: the d-default e applies to s iff there is no d' ⊇ s such that nπd' ⊆ d and the normal d'-worlds all fail e. Simplified for a single default: e applies to s if the frame can be coherently refined by e at d (from s's perspective).