On the State Complexity of Semi

On the State Complexity of Semi-Quantum
Finite Automata
Shenggen Zheng1 (Timothy), Jozef Grusk๐‘Ž1 , Daowen Qi๐‘ข2
1 Faculty of Informatics, Masaryk University, Czech Republic
2 Department of Computer Science, Sun Yat-sen University, China
arXiv:1307.2499
2014/3/17
27
1
1.Introduction
2014/3/17
27
2
1.Introduction
๏ƒ˜ Peter Shorโ€™s algorithm (1994)
A polynomial time algorithm to factor
large numbers on a quantum
computer.
RSA ( public-key encryption)
2014/3/17
27
3
1.Introduction
๏ƒ˜ USA, China, Europe
๏ƒ˜ Google, D-Wave
2014/3/17
27
4
What is quantum computing
๏ƒ˜ Quantum mechanics + computing
2014/3/17
27
5
What is quantum computing
๏ƒ˜ QM=probability + minus signs
(by Scott Aaronson) Amplitude (complex number)
๐‘š๐‘’๐‘ ๐‘ข๐‘Ÿ๐‘’๐‘š๐‘’๐‘›๐‘ก
ฮฑ
ฮฑ2
Probability: ๐‘1 + ๐‘2 > 0
Amplitude: ๐›ผ1 + ๐›ผ2 = 0
2014/3/17
27
6
What is quantum computing
If quantum computers are so great, how
come they have not been built yet?
๏ƒ˜ They have โ€“ and they have proved that 15 = 3 × 5
(with high probability) .
๏ƒ˜ Perhaps 21 = 3 × 7 will be the next technological
milestone.
2014/3/17
27
7
Quantum finite automata
Quantum computer
Quantum finite
automata
2014/3/17
27
8
Semi-quantum finite automata
Read only tape head
Classical
Control
C
Unit
Quantum
Memory
C
0 1
1 0
1
0
=
0
1
1/2 1/3 1
1/2
=
1/2 2/3 0
1/2
cos ๐œƒ
sin ๐œƒ 1
cos ๐œƒ
=
โˆ’ sin ๐œƒ cos ๐œƒ 0
โˆ’ sin ๐œƒ
Deterministic transition function: special stochastic matrix
Probabilistic transition function: stochastic matrix
Quantum transition function: unitary matrix
๏ƒ˜ Two-way finite automata with quantum and classical states
2QCFA (Ambainis, 02โ€™TCS)
๏ƒ˜ 1QCFA ( Zheng, 2011)
2014/3/17
27
9
2.State succinctness for promise problem
Promise problem : A = ๐ด๐‘ฆ๐‘’๐‘ , ๐ด๐‘›๐‘œ , A๐‘ฆ๐‘’๐‘  โˆฉ ๐ด๐‘›๐‘œ = โˆ…
Language: L = (๐ด๐‘ฆ๐‘’๐‘ , ๐ด๐‘›๐‘œ ), where ๐ด๐‘ฆ๐‘’๐‘  โˆช ๐ด๐‘›๐‘œ = ฮฃ โˆ—
๐ฟ
Promise problem: ฮฃ = 0,1, # ,
๐ด๐ธ๐‘„ (๐‘›) = (๐ด๐‘ฆ๐‘’๐‘  ๐‘› , ๐ด๐‘›๐‘œ (๐‘›)), where
๐ด๐‘›๐‘œ
๐ด๐‘ฆ๐‘’๐‘ 
๐ด๐‘ฆ๐‘’๐‘  = {๐‘ฅ#๐‘ฆ|๐‘ฅ = ๐‘ฆ, ๐‘ฅ, ๐‘ฆ โˆˆ {0,1}๐‘› }
๐ด๐‘›๐‘œ = {๐‘ฅ#๐‘ฆ|๐‘ฅ โ‰  ๐‘ฆ, ๐‘ฅ, ๐‘ฆ โˆˆ {0,1}๐‘› , ๐ป ๐‘ฅ, ๐‘ฆ = ๐‘›/2}
๐ป ๐‘ฅ, ๐‘ฆ is the Hamming distance between x and y.
๏ƒ˜ Deutsch-Jozsa promise problem in query complexity
๏ƒ˜ In communication complexity (H. Buhrman et al, STOCโ€™98)
Exact quantum communication complexity: ๐‘‚(log ๐‘›)
Exact classical: ฮฉ(๐‘›)
๏ƒ˜ Our result State complexity: ๐‘‚(๐‘›) vs 2ฮฉ(๐‘›)
2014/3/17
27
10
State succinctness for promise problem
๏ƒ˜Idea:
ฮฑ1 + ฮฑ2 + โ‹ฏ
2
=0
โ€ฆ
2014/3/17
27
11
State succinctness for promise problem
An exact 1QCFA algorithm: ๐‘ฅ = ๐‘ฅ1 โ‹ฏ ๐‘ฅ๐‘› , ๐‘ฆ = ๐‘ฆ1 โ‹ฏ ๐‘ฆ๐‘›
๐‘ฅ1 โ‹ฏ ๐‘ฅ๐‘› # ๐‘ฆ1 โ‹ฏ ๐‘ฆ๐‘›
1
0
โ‹ฎ
0
๐‘ˆ๐‘ 
1
๐‘ˆ๐‘“ ๐‘›
2014/3/17
1
1 1
๐‘› โ‹ฎ
1
๐‘ˆ1,๐‘ฅ1
(โˆ’1)
1
1
โ‹ฎ
๐‘›
1
๐‘›
๐‘ฅ๐‘– +๐‘ฆ๐‘–
(โˆ’1)
๐‘–=1
โˆ—
โ‹ฎ
โˆ—
๐‘ฅ1
๐‘ฅ
โ‹ฏ
(โˆ’1)๐‘ฅ1
1 (โˆ’1)๐‘ฅ2
โ‹ฎ
๐‘›
(โˆ’1)๐‘ฅ๐‘›
๐‘ฆ
โ‹ฏ
(โˆ’1)๐‘ฅ1 +๐‘ฆ1
1 (โˆ’1)๐‘ฅ2 +๐‘ฆ2
โ‹ฎ
๐‘›
(โˆ’1)๐‘ฅ๐‘› +๐‘ฆ๐‘›
If w โˆˆ ๐ด๐‘ฆ๐‘’๐‘  , ๐‘กโ„Ž๐‘’๐‘› ๐‘ฅ = ๐‘ฆ,
2
๐‘›
1
๐‘ฅ
+๐‘ฆ
โˆ’1 ๐‘– ๐‘– = 1
๐‘›
๐‘–=1
If ๐‘ค โˆˆ ๐ด๐‘›๐‘œ , then H x, y = n/2
2
๐‘›
1
โˆ’1 ๐‘ฅ๐‘– +๐‘ฆ๐‘– = 0
๐‘›
๐‘–=1
27
12
State succinctness for promise problem
Theorem 1. The promise problem ๐ด๐ธ๐‘„ can be solved
exactly by a 1QCFA with ๐‘› quantum basis states and ๐‘‚ ๐‘›
classical states, whereas the sizes of the corresponding
1DFA are 2ฮฉ(๐‘›) .
๏ƒ˜ The deterministic communication complexity for ๐ด๐ธ๐‘„
is at least 0.007n (H. Buhrman et al, STOCโ€™98)
Therefore, the sizes of the corresponding 1DFA are
2ฮฉ(๐‘›) (Kushilevitz, Advances in Computers, โ€˜97).
2014/3/17
27
13
3.State succinctness for ๐ฟ ๐‘
Language ๐ฟ ๐‘ = ๐‘Ž๐‘˜๐‘ ๐‘˜ โˆˆ โ„ค+ }.
๏ƒ˜ 1DFA and 1NFA: ๐‘ states
๏ƒ˜ One-way quantum finite automata (1QFA): ๐‘๐‘œ๐‘™๐‘ฆ
1
๐œ€
log ๐‘
(Ambainis and Freivalds, FOCSโ€™98)
log ๐‘
๏ƒ˜1QFA: O( 3 ) (Bertoni et al., TCS โ€™06)
๐œ€
log 2๐‘
๏ƒ˜1QFA: O(4
) (Ambainis,Nahimovs, TCS โ€™09)
๐œ€
๐›ผ
๐›ผ
๐›ผ
๏ƒ˜2DFA,2NFA:๐‘1 1 +๐‘2 2 +โ‹ฏ + ๐‘๐‘  ๐‘  ,where p =
๐›ผ
๐›ผ
๐›ผ
๐‘1 1 ๐‘2 2 โ‹ฏ ๐‘๐‘  ๐‘ 
(prime factorization) (Mereghetti, Pighizzini, J.Autom. Lan. Com โ€˜00)
๏ƒ˜One-way quantum finite automaton with restart (1QFAR):
1
๐œ‹
7 states, running time ๐‘‚( ๐‘ ๐‘–๐‘›โˆ’2 ( )|๐‘ค|) (Yakaryilmaz, Cem Say,
DMTCS โ€˜10)
2014/3/17
๐œ€
๐‘
27
14
State succinctness for ๐ฟ ๐‘
2QCFA, two quantum states, several classical states,
Idea: wheel of fortune, ๐‘ squares,
|๐‘ž0 =
1
0
,|๐‘ž1 = , ๐‘ˆ๐‘ =
0
1
๐œ‹
๐‘
๐œ‹
sin
๐‘
cos
๐œ‹
๐‘
๐œ‹
๐‘
๐œ‹
cos
๐‘
โˆ’ sin
๏ƒ˜If ๐‘ค โˆˆ ๐ฟ ๐‘ , |๐‘ž =|๐‘ž0
๏ƒ˜If ๐‘ค โˆ‰ ๐ฟ ๐‘ , ๐‘ค = ๐‘›
|๐‘ž = (๐‘ˆ๐‘ ) ๐‘› |๐‘ž0 = cos
๐‘›๐œ‹
|๐‘ž0
๐‘
+sin
๐‘›๐œ‹
|๐‘ž1
๐‘
Probability to observe |๐‘ž1 is ๐‘ƒ๐‘Ÿ = ๐‘ ๐‘–๐‘›2
๐‘›๐œ‹
๐‘
>
4
๐‘2
๏ƒ˜ Repeat for ๐‘‚(๐‘2 ) times ?
2014/3/17
27
15
State succinctness for ๐ฟ ๐‘
๏ƒ˜ Amplification of probabilities in a finite automaton
If ๐‘ค โˆˆ ๐ฟ: ๐‘ƒ๐‘Ž = 1 and ๐‘ƒ๐‘Ÿ = 0
If ๐‘ค โˆ‰ ๐ฟ: ๐‘ƒ๐‘Ž = 1 โˆ’ ๐‘ƒ๐‘Ÿ and ๐‘ƒ๐‘Ÿ > ๐‘ก(๐‘)
If ๐‘ค โˆˆ ๐ฟ: ๐‘ƒ๐‘Ž = ๐œ€ × ๐‘ก(๐‘) and ๐‘ƒ๐‘Ÿ = 0
If ๐‘ค โˆ‰ ๐ฟ: ๐‘ƒ๐‘Ž = ๐œ€ × ๐‘ก(๐‘) and ๐‘ƒ๐‘Ÿ > ๐‘ก(๐‘)
If ๐‘ค โˆˆ ๐ฟ: ๐‘ƒ๐‘Ž๐‘๐‘ = 1
If ๐‘ค โˆ‰ ๐ฟ: ๐‘ƒ๐‘Ÿ๐‘’๐‘— > 1 โˆ’ ๐œ€
1
๐œ€
Expected repetitions: ×
2014/3/17
๐‘ค
1
๐‘ก ๐‘
27
16
State succinctness for ๐ฟ ๐‘
Idea: 2QCFA, two quantum states, several classical states,
wheel of fortune, ๐‘ squares,
One iteration:
|๐‘ž0 =
1
0
,|๐‘ž1 = , ๐‘ˆ๐‘ =
0
1
๐œ‹
๐‘
๐œ‹
sin
๐‘
|๐‘ž = (๐‘ˆ๐‘ ) ๐‘› |๐‘ž0 = cos
๐œ‹
๐‘
๐œ‹
cos
๐‘
cos
๏ƒ˜If ๐‘ค โˆˆ ๐ฟ ๐‘ , |๐‘ž =|๐‘ž0
๏ƒ˜If ๐‘ค โˆ‰ ๐ฟ ๐‘ , ๐‘ค = ๐‘›
โˆ’ sin
๐‘ƒ๐‘Ž =
๐‘›๐œ‹
|๐‘ž0
๐‘
๐œ‹
๐‘
4๐œ€
๐‘2
๐‘›๐œ‹
|๐‘ž1
๐‘
๐‘›๐œ‹
๐‘ ๐‘–๐‘›2
๐‘
+sin
Probability to observe |๐‘ž1 is ๐‘ƒ๐‘Ÿ =
>
4
๐‘2
Use ๐‘ˆ๐‘,๐œ€ to control the expected number of repetitions.
2014/3/17
27
17
State succinctness for ๐ฟ ๐‘
4๐œ€
๐‘2
If ๐‘ค โˆˆ ๐ฟ ๐‘ , |๐‘ž = |๐‘ž0 , ๐‘กโ„Ž๐‘’๐‘› ๐‘ƒ๐‘Ž =
and ๐‘ƒ๐‘Ÿ = 0
๐‘ค
Repeated ad infinitum, the accepting probability
(1 โˆ’ ๐‘ƒ๐‘Ž )๐‘– 1 โˆ’ ๐‘ƒ๐‘Ÿ ๐‘– ๐‘ƒ๐‘Ž = 1
๏ƒ˜If ๐‘ค โˆ‰ ๐ฟ ๐‘ , ๐‘ƒ๐‘Ÿ โ‰ฅ
๐‘–โ‰ฅ0
4
and ๐‘ƒ๐‘Ž
๐‘2
=
4๐œ€
๐‘2
Repeated ad infinitum, the rejecting probability
(1 โˆ’ ๐‘ƒ๐‘Ž )๐‘– 1 โˆ’ ๐‘ƒ๐‘Ÿ ๐‘– ๐‘ƒ๐‘Ÿ > 1 โˆ’ ๐œ€
๐‘–โ‰ฅ0
1
๐œ€
๏ƒ˜ Expected running time ๐‘‚( ๐‘2 |๐‘ค|)
2014/3/17
27
18
State succinctness for ๐ฟ ๐‘
๏ƒ˜ Using a lemma in (Dwork and Stockmeyer, SIAM J. Comput.
โ€™90), we can prove the following theorem.
For any integer ๐‘, any polynomial expected running
time 2PFA recognizing ๐ฟ ๐‘ with error probability
๐œ€ < 1/2 has at least 3 (log ๐‘)/๐‘ states, where b is a
constant.
2014/3/17
27
19
4.State succinctness for C m
Language ๐ถ ๐‘š = {๐‘ค|๐‘ค โˆˆ ฮฃ ๐‘š }.
๏ƒ˜ DFA: ๐‘š
๏ƒ˜ 1PFA: ๐‘‚(log 2 ๐‘š) (Freivalds, Automatic Control and Computer
Sciences, โ€˜82)
๏ƒ˜ 1QFA: ๐‘‚(log ๐‘š) (Ambainis and Freivalds, FOCSโ€™98)
๏ƒ˜ 1QFAR: 7 quantum states, expected running time:
1 ๐‘š
๐‘‚ 2 |๐‘ค| , exponential of ๐‘š
๐œ€
(Yakaryilmaz and Cem Say, DMTCSโ€™10)
Our result: 2QCFA, 2 quantum states, several classical
1
states, expected running time: ๐‘‚ ๐‘š2 ๐‘ค 4 , polynomial of
๐œ–
๐‘š and |๐‘ค|.
2014/3/17
27
20
State succinctness for C m
Idea: |๐‘ž0 =
1
0
,|๐‘ž1 = ,
0
1
cos ๐‘š 2๐œ‹ sin ๐‘š 2 ๐œ‹
โˆ’ sin ๐‘š 2 ๐œ‹ cos ๐‘š 2๐œ‹
(rotated by the angle โˆ’ ๐‘š 2 ๐œ‹)
๐‘ˆ¢ =
๐‘ˆ๐œŽ = cos 2๐œ‹ โˆ’sin 2 ๐œ‹
sin 2 ๐œ‹ cos 2๐œ‹
(rotated by the angle 2๐œ‹)
๏ƒ˜ ๐‘› = |๐‘ค|,
2014/3/17
๐‘›โˆ’๐‘š
2
27
21
State succinctness for C m
If ๐‘ค โˆˆ ๐ถ(๐‘š), ๐‘ž = ๐‘ˆ๐›ผ
If w โˆ‰ ๐ถ ๐‘š , ๐‘ค = ๐‘›,
๐‘ž = ๐‘ˆ๐›ผ
๏‚ง
๐‘š
๐‘š
๐‘ˆ¢ ๐‘ž0 โŒช = ๐‘ž0 .
cos(๐‘› โˆ’ ๐‘š) 2๐œ‹
๐‘ˆ¢ ๐‘ž0 โŒช=
โˆ’ sin(๐‘› โˆ’ ๐‘š) 2 ๐œ‹
Pr =
๐œ€
2๐‘š2 ๐‘› + 1
2
sin(๐‘› โˆ’ ๐‘š) 2 ๐œ‹
|๐‘ž0 โŒช
cos(๐‘› โˆ’ ๐‘š) 2๐œ‹
๐‘ค
Probability to get |๐‘ž1 is
sin2 (๐‘›
๐‘ƒ๐‘Ž =
1
โˆ’ ๐‘š) 2 ๐œ‹ >
2 ๐‘šโˆ’๐‘›
2
+1
(see Ambainis and Watrous, TCS โ€™02; Zheng et al. TCS โ€™13)
Use ๐‘ˆ๐‘š,๐œ€ and two random walks to control the expected number of
repetitions.
2014/3/17
27
22
State succinctness for C m
If ๐‘ค โˆˆ ๐ถ ๐‘š , |๐‘ž = |๐‘ž0 , ๐‘กโ„Ž๐‘’๐‘› ๐‘ƒ๐‘Ž =
๐œ€
2๐‘š2 ๐‘›+1 2
and ๐‘ƒ๐‘Ÿ = 0
Repeated for ad infinitum, the accepting probability
(1 โˆ’ ๐‘ƒ๐‘Ž )๐‘– 1 โˆ’ ๐‘ƒ๐‘Ÿ ๐‘– ๐‘ƒ๐‘Ž = 1
๐‘–โ‰ฅ0
๏ƒ˜If ๐‘ค โˆ‰ ๐ถ ๐‘š , ๐‘ƒ๐‘Ÿ >
1
2 ๐‘šโˆ’๐‘› 2 +1
and ๐‘ƒ๐‘Ž <
๐œ€
2๐‘š2 ๐‘›+1 2
Repeated for ad infinitum, the rejecting probability
(1 โˆ’ ๐‘ƒ๐‘Ž )๐‘– 1 โˆ’ ๐‘ƒ๐‘Ÿ ๐‘– ๐‘ƒ๐‘Ÿ > 1 โˆ’ ๐œ€
๐‘–โ‰ฅ0
๏ƒ˜ Expected running time ๐‘‚
2014/3/17
1 2
๐‘š
๐œ–
27
๐‘ค
4
.
23
State succinctness for C m
๏ƒ˜ Using a lemma in (Dwork and Stockmeyer, SIAMJ. Comput.
โ€™90), we can prove the following theorem.
For any integer ๐‘, any polynomial expected running
time 2PFA recognizing C ๐‘š with error probability
๐œ€ < 1/2 has at least 3 (log ๐‘š)/๐‘ states, where b is a
constant.
2014/3/17
27
24
5.A trade-off property of 1QCFA
๏ƒ˜ Quantum resources are expensive and hard to deal
with. One can expect to have only very limited
number of qubits in current quantum system.
๏ƒ˜ ๐‘ = ๐‘ž1 × ๐‘ž2 and
gcd ๐‘ž1 , ๐‘ž2 = 1.
๐ฟ ๐‘ = ๐ฟ ๐‘ž1 โˆฉ ๐ฟ(๐‘ž2 ) ,
๏ƒ˜ ๐ฟ ๐‘ž1 1QCFA ฮ‘(๐‘ž1 )
simulated by
๏ƒ˜ ๐ฟ ๐‘ž2 1DFA
1QCFA ฮ‘(๐‘ž2 )
๏ƒ˜ 1QCFA is closed under intersection
State complexity: ๐‘‚ log ๐‘ž1 + ๐‘‚(๐‘ž2 )
2014/3/17
27
25
6. Concluding remarks
๏ƒ˜ We have proved exact 1QCFA state succinctness for a
promise problem.
๏ƒ˜ We have proved 2QCFA state succinctness for two
languages
๏ƒ˜ A trade-off property of 1QCFA
๏ƒ˜ Future work
It has been proved (Klauck, STOC โ€˜00) that exact 1QCFA have
not state complexity advantage over 1DFA in recognizing a
language. How about exact 2QCFA vs. 2DFA?
2014/3/17
27
26
2014/3/17
27
27