Table of Contents
Fetching ...

Classical and quantum Merlin-Arthur automata

Abuzer Yakaryılmaz

TL;DR

This work studies Merlin-Arthur automata in which a certificate is supplied by a prover before the input is read, and analyzes how certificate length and postselection affect verification power across deterministic, probabilistic, and quantum finite-state verifiers. It defines MA-DFAs, MA-PFAs, MA-QFAs and their postselecting variants, and systematically maps certificate-size regimes—from sublinear to exponential to arbitrarily long—to the classes of languages they can verify on unary and binary inputs. A key finding is that MA-DFAs with constant certificates recognize exactly the regular languages (via equivalence with multi-entry DFAs), while longer certificates empower MA-PFAs and MA-QFAs to verify nonregular and even all decidable/universal binary languages under appropriate conditions; postselection further amplifies these capabilities, especially for quantum verifiers. Overall, the paper delineates a spectrum of certificate-size versus power in automata augmented with pre-input certificates, connecting classical automata, interactive proofs, and quantum verification concepts with concrete results for unary and binary languages.

Abstract

We introduce Merlin-Arthur (MA) automata where Merlin provides a certificate at the beginning of computation and it is scanned by Arthur before reading the input. We define Merlin-Arthur deterministic, probabilistic, and quantum finite state automata (resp., MA-DFAs, MA-PFAs, and MA-QFAs) and postselecting MA-PFAs and MA-QFAs (resp., MA-PostPFA and MA-PostQFA). We present several results using different certificate lengths. We show that MA-DFAs use constant length certificates, and they are equivalent to multi-entry DFAs. Thus, they recognize all and only regular languages, but they can be exponential and polynomial state efficient over binary and unary languages, respectively. With sublinear length certificates, MA-PFAs can recognize several nonstochastic unary languages with cutpoint 1/2. With linear length certificates, MA-PostPFAs can recognize these nonstochastic unary languages with bounded error. With arbitrarily long certificates, bounded-error MA-PostPFAs can verify every unary decidable language. With sublinear length certificates, bounded-error MA-PostQFAs can verify several nonstochastic unary languages. With linear length certificates, they can verify every unary language and some NP-complete binary languages. With exponential length certificates, they can verify every binary language.

Classical and quantum Merlin-Arthur automata

TL;DR

This work studies Merlin-Arthur automata in which a certificate is supplied by a prover before the input is read, and analyzes how certificate length and postselection affect verification power across deterministic, probabilistic, and quantum finite-state verifiers. It defines MA-DFAs, MA-PFAs, MA-QFAs and their postselecting variants, and systematically maps certificate-size regimes—from sublinear to exponential to arbitrarily long—to the classes of languages they can verify on unary and binary inputs. A key finding is that MA-DFAs with constant certificates recognize exactly the regular languages (via equivalence with multi-entry DFAs), while longer certificates empower MA-PFAs and MA-QFAs to verify nonregular and even all decidable/universal binary languages under appropriate conditions; postselection further amplifies these capabilities, especially for quantum verifiers. Overall, the paper delineates a spectrum of certificate-size versus power in automata augmented with pre-input certificates, connecting classical automata, interactive proofs, and quantum verification concepts with concrete results for unary and binary languages.

Abstract

We introduce Merlin-Arthur (MA) automata where Merlin provides a certificate at the beginning of computation and it is scanned by Arthur before reading the input. We define Merlin-Arthur deterministic, probabilistic, and quantum finite state automata (resp., MA-DFAs, MA-PFAs, and MA-QFAs) and postselecting MA-PFAs and MA-QFAs (resp., MA-PostPFA and MA-PostQFA). We present several results using different certificate lengths. We show that MA-DFAs use constant length certificates, and they are equivalent to multi-entry DFAs. Thus, they recognize all and only regular languages, but they can be exponential and polynomial state efficient over binary and unary languages, respectively. With sublinear length certificates, MA-PFAs can recognize several nonstochastic unary languages with cutpoint 1/2. With linear length certificates, MA-PostPFAs can recognize these nonstochastic unary languages with bounded error. With arbitrarily long certificates, bounded-error MA-PostPFAs can verify every unary decidable language. With sublinear length certificates, bounded-error MA-PostQFAs can verify several nonstochastic unary languages. With linear length certificates, they can verify every unary language and some NP-complete binary languages. With exponential length certificates, they can verify every binary language.
Paper Structure (13 sections, 16 theorems, 87 equations)

This paper contains 13 sections, 16 theorems, 87 equations.

Key Result

Theorem 1

If a language $L$ is verified by an $m$-state MA-DFA, then $L$ is recognized by some $k$-entry DFAs with $m$ states, where $k \leq m$.

Theorems & Definitions (24)

  • Theorem 1
  • Theorem 2
  • proof
  • Corollary 3
  • Theorem 4
  • proof
  • Corollary 5
  • Corollary 6
  • Theorem 7
  • proof
  • ...and 14 more