BARCELONA (UAB) THE UNIVERSITY OF TOKYO (UT) WITH C. HIRCHE (UAB), M. KOASHI (UT), AND A. WINTER(UAB) PRX 7, 021006 (2017) (SEE ALSO ARXIV:1609.07021) Slides are available at: https://sites.google.com/view/yoshifuminakata/home
designs in quantum information science? • Main results – new quantum circuits for unitary designs – 2. Constructions of designs • Quantum circuits • Combinatorics – local permutation check – • Hamiltonian construction – Sherrington-Kirkpartick spinglass – 3. Conclusion and outlook
random unitary is the unique unitarily invariant probability measure H on the unitary group (). (A) H( ) = 1, (B) for any ∈ () and any (Borel) set ⊆ , H = H = H . A uniform distribution of the Haar measure Unitary group
random unitary is the most important tool in quantum information science. In Quantum Information Processing • Quantum communication – Random encoding to achieve the quantum capacity [Hayden et al ‘07] • Quantum computation – To show quantum computational supremacy [Mann & Bremner ‘17] • Experiments of quantum devices – Randomized benchmarking [Knill et.al ‘08] – Calibration of multi-qubit Q.circuits [Boixo ‘16] • Q. encryption, Q. algorithms, etc… In fundamental physics • Disordered many-body systems – Random matrix theory [Metha ‘04] • Foundation of quantum statistical mechanics – Thermalization in isolated systems [Popescu et al ‘07] – Pre-thermalization [Reimann ‘16] • Quantum black holes – Solving the information paradox [Hayden & Preskill ‘07] – Holographic codes [Patawskiet al ‘15] • Quantum chaos – Out-of-time-ordered correlators [Roberts & Yoshida ‘16]
random unitary is the most important tool in quantum information science. • An -qubit system is described by a Hilbert space ℋ = (ℂ2)⨂. – In QIP, we need Haar random unitaries on (2)…. • Can we generate Haar random unitaries on such a large dimension? – In QIP, we use quantum circuits. The basic unit is a qubit. = a unit vector in ℂ2. Exponentially large in
Quantum gate on qubits is a unitary ∈ 2 , where is independent of . A HAAR RANDOM UNITARY IN QIP 3 2nd qubit 3rd qubit 1 ∈ (22) th qubit ( − 1)th qubit Time 1st qubit 2 ∈ (23) 3 4 5 1 = 1 ⨂ 2−2×2−2 ∈ (2) 2 = 2×2 ⨂2 ⨂ 2−4×2−4 ∈ (2) circuit = … 2 1 ∈ (2) (: # of gates in the circuit) When = poly(), the circuit is efficient. ℋ = (ℂ2)⨂
gates are needed to implement Haar random unitaries on N qubits? – At least ( ) quantum gates, which is extremely inefficient!! – We can never use a Haar random unitary in practice… • In QIP, a random unitary approximating Haar random unitaries is enough ☺ – A unitary -design comes into the play in QIP! 2nd qubit 3rd qubit th qubit ( − 1)th qubit 1st qubit 3 4 5 circuit = … 2 1 ∈ (2) (: # of gates in the circuit)
Can we generate approximate unitary -designs on qubits efficiently by Q. circuits? − Especially when is large and is small (e.g. = 2,4). A distribution of the Haar measure Unitary group A distribution of a unitary design Unitary group ・ ・ ・ ・ ・ ・ ・ ・ ・・ To simulate only low order moments : one unitary Enough for quantum information processing! Approximate unitary design An -approximate unitary -design is a probability measure , that simulates up to the th order statistical moments of the Haar measure within an error .
et al ‘01, Cleve et al ’15 etc] – Due to the fact that the Clifford group forms a unitary 2-design. • Only two constructions were known for -designs ( ≥ 3) on qubits Harrow and Low '09 Brandao et al '13 Our result Quantum circuit Q. Fourier Transformation + Toffoli-type gates Local random circuits Hadamard gates + random diagonal gates Methods Markov chain Gap problem of many-body Hamiltonian Combinatorics # of gates (33) [Brodsky & Hoory ’13] (102) Θ(2) Works for = (/ log ) ≤ poly() = ( )
'13 Our result Quantum circuit Q. Fourier Transformation + Toffoli-type gates Local random circuits Hadamard gates + random diagonal gates Methods Markov chain Gap problem of many-body Hamiltonian Combinatorics # of gates (33) [Brodsky & Hoory ’13] (102) Θ(2) Works for = (/ log ) = (poly ) = ( ) • Many constructions of unitary 2-designs [DiVincenzo et al ‘01, Cleve et al ’15 etc] – Due to the fact that the Clifford group forms a unitary 2-design. • Only two constructions were known for -designs ( ≥ 3) on qubits
'13 Our result ‘16 Quantum circuit Q. Fourier Transformation + Toffoli-type gates Local random circuits Hadamard gates + random diagonal gates Methods Markov chain Gap problem of many-body Hamiltonian Combinatorics # of gates (33) [Brodsky & Hoory ’13] (102) Θ(2) Works for = (/ log ) = (poly ) = ( ) • Many constructions of unitary 2-designs [DiVincenzo et al ‘01, Cleve et al ’15 etc] – Due to the fact that the Clifford group forms a unitary 2-design. • Only two constructions were known for -designs ( ≥ 3) on qubits
circuits with Θ(2) gates generating a unitary -design. • Main Theorem = Theorem 1 + Theorem 2 Theorem 1 (informal, Proof skipped) Let and be bases in ℋ satisfying a Fourier-type condition. Define D and D by = diag 1, … , , and = diag (1, … , ), where all phases are random. Then, repeating and D O() times becomes a unitary t-design. Unitary -design 1 2 1 2 However, no efficient quantum circuit for (as = 2, too many phases to be randomized).
circuits with Θ(2) gates generating a unitary -design. • Main Theorem = Theorem 1 + Theorem 2 Theorem 2 (informal) Let be the Pauli-Z basis. Then, can be simulated up to the th order by random diagonal-circuits (RDC()) with Θ(2) gates. Unitary -design 1 2 1 2
circuits with Θ(2) gates generating a unitary -design. • Main Theorem = Theorem 1 + Theorem 2 RDC(S) RDC(S) 1 2 Unitary -design Remark 1: When = Pauli-Z basis, = the Pauli-X basis: = ⨂⨂. Theorem 2 (informal) Let be the Pauli-Z basis. Then, can be simulated up to the th order by random diagonal-circuits (RDC()) with Θ(2) gates.
circuits with Θ(2) gates generating a unitary -design. • Main Theorem = Theorem 1 + Theorem 2 RDC(ℑ) RDC(ℑ) RDC(ℑ) RDC(ℑ) Unitary -design Repeating RDC() and the Hadamard gates is a unitary t-design! Remark 1: When = Pauli-Z basis, = the Pauli-X basis: = ⨂⨂. Theorem 2 (informal) Let be the Pauli-Z basis. Then, can be simulated up to the th order by random diagonal-circuits (RDC()) with Θ(2) gates.
a subset with a constant # of elements, and be {1 , 2 , … , }. A RDC() is a quantum circuit as follows: RANDOM DIAGONAL CIRCUITS 1 Diagonal in the Pauli-Z basis RDC(S) 1 2 3 4 5 diag 1, 2, … … … All phases are randomly & independently chosen from [0, 2). diag 1, 2, … … … E.g.) S={1 , 2 , 3 , 4 }, where 1 = 1,2 , 2 = 2,4,5 , 3 = 3,5 , and 4 = 1,3,5 .
RDC() simulate = diag (1, … , 2 ) to the th order? – Equivalent to the local permutation check problem. Diagonal in the Pauli-Z basis RDC(S) ≈ Up to the th order moments diag (, … , )
’) with each entry {0, 1}. • For a given S={1 , 2 , … , }, (1): the rows of K are NOT permutation of the rows of K’. (2): for every , the rows of are the permutation of ′ . 0 1 1 0 1 0 1 1 1 1 0 0 1 1 1 0 1 1 0 1 0 1 1 1 1 1 0 1 1 0 = ′ = = 3 & = 5 Λ S ≔ {(, ′): (, ′) satisfies the conditions (1) and (2). }
’) with each entry {0, 1}. • For a given S={1 , 2 , … , }, Λ S ≔ {(, ′): (, ′) satisfies the conditions (1) and (2). } (1): the rows of K are NOT permutation of the rows of K’. (2): for every , the rows of are the permutation of ′ . 0 1 1 0 1 0 1 1 1 1 0 0 1 1 1 0 1 1 0 1 0 1 1 1 1 1 0 1 1 0 = ′ = = 3 & = 5 S = 1 , 2 1 = {1,2}, 2 = {2,4,5} 1 ′1 2 ′2 (, ′) ∈ Λ S
’) with each entry {0, 1}. • For a given S={1 , 2 , … , }, Λ S ≔ {(, ′): (, ′) satisfies the conditions (1) and (2). } (1): the rows of K are NOT permutation of the rows of K’. (2): for every , the rows of are the permutation of ′ . 0 1 1 0 1 0 1 1 1 1 0 0 1 1 1 0 1 1 0 1 0 1 1 1 1 1 0 1 1 0 = ′ = = 3 & = 5 S = 1 , 2 1 = {1,2}, 2 = {2,4,5} 1 ′1 2 ′2 (, ′) ∈ Λ S Local permutation check For a given S = {1 , 2 , … , }, count the # of (K, K′) in Λ S .
given S = {1 , 2 , … , }, count the # of (K, K′) in Λ S ( = |Λ S | ). • For what S, does the RDC() simulate = diag (1, … , 2 ) to the th order? – Equivalent to the local permutation check problem. Diagonal in the Pauli-Z basis RDC(S) ≈ Up to the th order moments diag (, … , )
given S = {1 , 2 , … , }, count the # of (K, K′) in Λ S ( = |Λ S | ). Lemma: For a given S = {1 , … , }, if |Λ S | = O(2 −1 ), then RDC() simulates = diag (1, … , ) up to the th order. Find = {1 , 2 , … , } such that 1. | | = O( − ) 2. (# of gates in the circuit) is as small as possible. • For what S, does the RDC() simulate = diag (1, … , 2 ) to the th order? – Equivalent to the local permutation check problem.
found a good choice: Sp = , : ≠ , , ∈ {1, … , } . – The most non-trivial in the whole analysis. LPC VS UNITARY DESIGN 3 Lemma: |Λ(p)| ≤ 2 t−1 N+2t2 , and Sp = = Θ(2). Find = {1 , 2 , … , } such that 1. | | = O( − ) 2. (# of gates in the circuit) is as small as possible. Theorem 2 The can be simulated up to the th order by RDC(p) with Θ(2) gates.
If and satisfy a Fourier-type condition, repeating and D O() times becomes a unitary t-design. Unitary -design 1 2 1 2 Theorem 2 The can be simulated up to the th order by RDC(p) with Θ(2) gates.
simulated up to the th order by RDC(p) with Θ(2) gates. RDC(Sp ) Unitary -design RDC(Sp ) RDC(Sp ) RDC(Sp ) Theorem 1 (informal, Proof skipped) If and satisfy a Fourier-type condition, repeating and D O() times becomes a unitary t-design.
spin-glass Hamiltonian, Then, the time evolution operators generated by () forms a unitary -design after th ≈ 2. Main theorem Suppose that = ( ). Then, the following quantum circuit forms an -approximate unitary -design, where we use Θ(2 + log2 (1/))) two-qubit gates.
2. A local permutation check problem is proposed. – Found one specific solution. 3. Point out that the Sherrington-Kirkpatrick spin-glass Hamiltonian ↔ unitary design. – Toy model for OTOC and quantum duality? Harrow and Low ’09 Brandao et al ‘13 Our result ‘16 Methods Markov chain Many-body Hamiltonian Combinatorics Length of the circuit (33) [Brodsky & Hoory ’13] (102) Θ(2) Works for = (/ log ) = (poly ) = ( )
May result in more efficient quantum circuits. • Lower bounds on the # of quantum gates to generate unitary designs? – Trivial bound is (), possible to show (log). • Efficient quantum circuits for exact ( = 0) unitary -designs on qubits? − None is known for ≥ 4. − From the Carathéodory's theorem, there exist exact designs. THANK YOU FOR YOUR ATTENTION! PRX 7, 021006 (2017) (SEE ALSO ARXIV:1609.07021) Slides are available at: https://sites.google.com/view/yoshifuminakata/home