The output tailored verifier of the introspection presentation, from its specification #
The output tailored verifier at λ has seven's sampler and two programs, an answer-length
calculator L and a linear-constraints processor P. At an index n they meet the
specification IntroSpec when, at every question vectorEquiv x of the detyped presentation,
L outputs the presented game's lengths and P its constraints. Then the output's n-th game
extends the presented one along vectorEquiv (extends_presented), so it has its value and its
perfect ZPC strategies.
On the other side, the detyped game the presentation is compared with is the reference
verifier's game read along vectorEquiv (quantumValue_H), whose value is the value of
seven's output (output_val_eq_reference). So the output's val* is at most the value of
seven's output at the same index (valStar_output_le).
The reference verifier at index n.
Equations
- MIPRE.Tailored.Intro.Output.Ref c hc he lam n U V = MIPRE.Introspection.DecisionCompiler.reference c hc he U (V.sampler.prog, V.decider.prog) lam n
Instances For
The detyped game is the reference verifier's game, read along vectorEquiv.
The presentation's value is at most that of seven's output, on the input's normal
form verifier.
The output tailored verifier and its specification #
The output tailored verifier with programs L and P: seven's sampler.
Equations
- MIPRE.Tailored.Intro.Output.outTV c hc he lam L P = { sampler := MIPRE.Introspection.PauliSampler.finalSampler c hc he 7 lam, len := L, lp := P }
Instances For
The detyped questions, as the output's questions.
Equations
- MIPRE.Tailored.Intro.Output.qe c hc he U V hl hn = (MIPRE.CL.Detyping.DeciderProgram.vectorEquiv (MIPRE.Introspection.PauliSampler.dimension c lam n)).trans (MIPRE.Verifier.dimensionEquiv ⋯)
Instances For
The specification of the output's two programs at index n: at every detyped question
the answer-length calculator outputs the presented game's lengths, and the linear-constraints
processor its constraints.
- lenR_eq (x : CL.Detyping.Coord Introspection.DecisionKernel.Label (Fin (Introspection.PauliSampler.dimension c lam n)) → ZMod 2) : LenIs L n (CL.toBits ((qe c hc he U V hl hn) x)) false (G.lenR x)
- lenL_eq (x : CL.Detyping.Coord Introspection.DecisionKernel.Label (Fin (Introspection.PauliSampler.dimension c lam n)) → ZMod 2) : LenIs L n (CL.toBits ((qe c hc he U V hl hn) x)) true (G.lenL x)
- cons_eq (x y : CL.Detyping.Coord Introspection.DecisionKernel.Label (Fin (Introspection.PauliSampler.dimension c lam n)) → ZMod 2) (aR bR : Cost.BitStr) : LpIs P n (CL.toBits ((qe c hc he U V hl hn) x)) (CL.toBits ((qe c hc he U V hl hn) y)) aR bR (G.cons x y aR bR)
Instances For
The output's game extends the presented game along the detyped questions.
Soundness #
The output's val* is at most the value of seven's output.
Soundness of the output, through seven.
Completeness #
Completeness of the output: a perfect PCC strategy of the reference verifier's typed game
meeting the three conditions of Complete.hasPerfectZPC_tpresented gives a perfect ZPC strategy
of the output's n-th game.
The input's normal form verifier #
The input's normal form verifier accepts as the input does.
The loss #
Soundness in the contract's form #
Soundness of the output, in the contract's form: for a λ-bounded input whose normal
form verifier is Cλ-bounded (C ≥ 2), when the output's programs meet the specification at
Cλ, val* of the output above 1 - ε gives val* of the input at 2^n at least
1 - δ(a C^a, b, λ, n, ε), with seven's a and b.