Asymptotic performance of port-based teleportation - arXiv ...
Transcript of Asymptotic performance of port-based teleportation - arXiv ...
Asympto c performance of
port-based teleporta on
arXiv:1809.10751
Felix Leditzky
(JILA & CTQM, University of Colorado Boulder)
Joint work with
M. Christandl, C. Majenz, G. Smith, F. Speelman, M. Walter
BIRS workshop, Banff, Canada
26 July 2019
Resource conversion in teleporta on
▶ Standard teleporta on: [Benne et al. 1993]
[qq] + 2[c → c] ≥ [q → q]
▶ Port-based teleporta on: [Ishizaka and Hiroshima 2008]
N[qq] + (logN)[c → c] ≥ [q → q]
This is worse... why care?
Because port-based teleporta on has unitary covariance!
2 / 35
Resource conversion in teleporta on
▶ Standard teleporta on: [Benne et al. 1993]
[qq] + 2[c → c] ≥ [q → q]
▶ Port-based teleporta on: [Ishizaka and Hiroshima 2008]
N[qq] + (logN)[c → c] ≥ [q → q]
This is worse... why care?
Because port-based teleporta on has unitary covariance!
2 / 35
Resource conversion in teleporta on
▶ Standard teleporta on: [Benne et al. 1993]
[qq] + 2[c → c] ≥ [q → q]
▶ Port-based teleporta on: [Ishizaka and Hiroshima 2008]
N[qq] + (logN)[c → c] ≥ [q → q]
This is worse... why care?
Because port-based teleporta on has unitary covariance!
2 / 35
Standard teleporta on protocol
Alice Bob
A0
A1 B1
3 / 35
Standard teleporta on protocol
Alice Bob
A0
A1 B1
3 / 35
Standard teleporta on protocol
Alice Bob
m
3 / 35
Standard teleporta on protocol
Alice Bob
mUm
3 / 35
Standard teleporta on protocol
Alice Bob
mUm
B0
3 / 35
Standard teleporta on protocol
Alice Bob
mUm
B0
3 / 35
Port-based teleporta on
Alice BobA0
A1 B1
Ai Bi
AN BN
[Ishizaka and Hiroshima 2008]
4 / 35
Port-based teleporta on
Alice BobA0
A1 B1
Ai Bi
AN BN
4 / 35
Port-based teleporta on
Alice Bob
i
4 / 35
Port-based teleporta on
Alice Bob
i
Tr1...i‐1,i+1...N
4 / 35
Port-based teleporta on
Alice Bob
B0
i
Tr1...i‐1,i+1...N
4 / 35
Port-based teleporta on
Alice BobA0
A1 B1
Ai Bi
AN BN
i
Tr1...i‐1,i+1...N
[Ishizaka and Hiroshima 2008]
5 / 35
Port-based teleporta on
Alice BobA0
A1 B1
Ai Bi
AN BN
i
Tr1...i‐1,i+1...N
[Ishizaka and Hiroshima 2008]
5 / 35
Why is PBT interes ng?
▶ Par al trace commutes withW⊗N: PBT is unitarily covariant.
▶ PBT enables instantaneous non-local quantum computa on
(INQC). [Beigi and König 2011]
▶ INQC can be used to break posi on-based cryptography.
[Buhrman et al. 2014]
Caveat
Unitary covariance leads to the fact that
perfect PBT is impossible with finite resources.
[Nielsen and Chuang 1997; Ishizaka and Hiroshima 2008]
6 / 35
Variants of PBT
Determinis c PBT
Protocol always yields final state that approximates target state.
Probabilis c PBT
Protocol yields exact target state with certain probability.
▶ Unitary covariance: Perfect PBT impossible with finite
resources.
▶ Goal of this talk: Understand symmetries of PBT and determine
asympto c performance of PBT protocols.
7 / 35
Outline
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
8 / 35
Table of Contents
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
9 / 35
Determinis c PBT
▶ In determinis c PBT the protocol always yields a final state as an
approxima on to the target state.
▶ Hence, PBT protocol implements qudit channel Λ that
simulates ideal channel.
▶ Figure of merit: entanglement fidelity
Fd = F(Λ, id) = 〈Φ+A′A|(id⊗Λ)(Φ+
A′A)|Φ+A′A〉
▶ For PBT the diamond norm distance is exactly equivalent to Fd:
‖ id−Λ‖⋄ = 2(1− Fd). [Pirandola et al. 2018]
10 / 35
Determinis c PBT and state discrimina on
▶ Determinis c PBT is equivalent to state discrimina on of the
uniformly drawn states [Ishizaka and Hiroshima 2009]
ω(i)ANB = TrBic φANBN .
▶ Success probability q of discrimina ng between ω(i):
q =d2
NFd.
▶ Suggests pre y good measurement (PGM) as POVM.
▶ Further protocol simplifica on: |φ〉 = EPR⊗N
▶ We call (EPR⊗N, PGM) the standard protocol.
11 / 35
Probabilis c PBT
▶ Probabilis c PBT yields the exact target state with success
probability pd and aborts otherwise.
▶ Extended POVM Eprob = {E(i)}Ni=0, where E(0) corresponds to
abor on of the protocol.
▶ Probabilis c PBT is a special case of determinis c PBT.
(Send random port when ge ng outcome ”0”.)
▶ Again: consider special case where |φ〉 = EPR⊗N.
▶ We call (EPR⊗N, Eprob) the EPR protocol.
(POVM Eprob is now op mized over.)
12 / 35
Exis ng results: op mal performance of PBT
▶ Standard determinis c protocol:
Fstdd ≥ 1− d2 − 1N
.
[Ishizaka and Hiroshima 2008; Beigi and König 2011]
▶ Converse bound for arbitrary determinis c protocols:
F∗d ≤ 1− 14(d− 1)N2 + O(N−3). [Ishizaka 2015]
▶ Closed forms for d = 2: [Ishizaka and Hiroshima 2009]
Fstd2 = FEPR2 = 1− 34N
+ o(N−1)
pEPR2 ∼ 1−( 8πN
)−1/2+ o(N−1/2).
13 / 35
Exis ng results: op mal performance of PBT
▶ PBT has a lot of inherent symmetries
−→ use representa on theory (RT)!
▶ Leads to exact expressions for Fd and pd in terms of RT
quan es. [Studziński et al. 2017] and [Mozrzymas et al. 2017]
▶ Our main results: Asympto cs of these expressions for F∗d, Fstdd
and pEPRd to first order.
▶ This talk focuses on
▷ standard determinis c protocol Fstdd ;
▷ EPR probabilis c protocol pEPRd .
14 / 35
Table of Contents
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
15 / 35
Natural symmetries of PBT
Nota on: Ud … unitary group; SN … symmetric group.
Permuta on symmetry
Every port Bi is equally good for teleporta on
−→ SN-symmetry of ρBN = TrAN φANBN .
Same symmetry for POVM elements
−→ SN-ac on on {E(i)}i.
Unitary invariance
The protocol works equally well for all input states
−→ Ud-symmetry of ρBN .
16 / 35
Natural symmetries of PBT
Proposi on: Symmetries of PBT
Every PBT protocol Λ can be symmetrized to a protocol Λs with
F(Λs, id) ≥ F(Λ, id), sa sfying:
▶ Resource state φANBN is a purifica on of a symmetric Werner
state, i.e., invariant under U⊗NA ⊗ U⊗N
B and SN.
▶ SN acts on {E(i)}i, and each E(i) is invariant under UA0 ⊗ U⊗NA .
▶ Λs is unitarily covariant.
”Folklore” results, proofs in C. Majenz’s PhD thesis and our paper.
17 / 35
Schur-Weyl duality
▶ Resource state φANBN invariant under ac on of Ud SN
−→ structure determined by Schur-Weyl duality.
▶ Group ac ons:
|ψ1〉 ⊗ . . .⊗ |ψN〉π∈SN7−−−−→ |ψπ−1(1)〉 ⊗ . . .⊗ |ψπ−1(N)〉
|ψ1〉 ⊗ . . .⊗ |ψN〉U∈Ud7−−−−→ U|ψ1〉 ⊗ . . .⊗ U|ψN〉
▶ Schur-Weyl decomposi on:
(Cd)⊗N ∼=⊕λ⊢d N
[λ]⊗ Vλ
▶ λ `d N: Young diagram with N boxes and at most d rows.
▶ Irreducible representa ons:
▷ [λ] is an irrep of SN with dim[λ] = dλ.
▷ Vλ is an irrep of Ud with dimVλ = md,λ.18 / 35
Exact expressions for Fd and pd using RT
[Studziński et al. 2017; Mozrzymas et al. 2017]
Standard determinis c protocol
Fstdd =1
dN−2
∑α ⊢d N−1
( ∑μ=α+□
√dμmd,μ
)2
,
where μ = α +□ denotes a Young diagram μ `d N obtained from
α `d N− 1 by adding a single box (!).
EPR probabilis c protocol
pEPRd =1dN
∑α ⊢d N−1
m2αdμ∗mμ∗
where μ∗ is the Young diagram obtained from α `d N− 1 by adding a single
box such that Nmd,μdαmαdμ
is maximal.
19 / 35
Table of Contents
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
20 / 35
Main results
Main result 1: standard determinis c PBT
For determinis c PBT using PGM and EPR pairs, we prove:
Fstdd = 1− d2 − 14N
+ O(N−3/2+δ) for any δ > 0.
▶ Recovers qubit result Fstd2 = 1− 34N
+ o(N−1).
▶ Shows that Fstdd ≥ 1− d2 − 1N
is not ght, confirming numerical
evidence.
21 / 35
Main results
Main result 2: probabilis c PBT
For probabilis c PBT using EPR, we prove:
pEPRd = 1−√
dN− 1
E[λmax(G)] + o(N−1),
where G is a Gaussian unitary, i.e, a Hermi an, traceless random
d× dmatrix with independent Gaussian RVs as entries.
▶ For qubits (i.e., d = 2 and G is a 2× 2 matrix):
E[λmax(G)] = 2π−1/2.
▶ Hence, our result ”corrects” the qubit result
pEPR2 ∼ 1−√
8πN + o(N−1/2). [Ishizaka and Hiroshima 2009]
▶ Arbitrary d: use bounds on E[λmax(G)].22 / 35
Standard determinis c protocol
0 50 100 150 200
0.4
0.6
0.8
1
N
FstddFstdd,asymptd = 2d = 3d = 4d = 5
Fstdd =1
dN−2
∑α ⊢d N−1
(∑μ=α+□
√dμmd,μ
)2Fstdd,asympt = 1− d2−1
4N
23 / 35
EPR probabilis c protocol
0 100 200 300 400 5000
0.2
0.4
0.6
0.8
1
N
pEPRd
pEPRd,asymptd = 2d = 3d = 4d = 5
pEPRd =1dN
∑α ⊢d N−1
m2αdμ∗mμ∗
pEPRd,asympt = 1− E[λmax(G)]√
dN−1
24 / 35
Table of Contents
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
25 / 35
Spectrum es ma on & random matrix theory
▶ Recall Schur-Weyl duality:
(Cd)⊗N ∼=⊕λ⊢d N
[λ]⊗ Vλ.
▶ Consider the projec ve measurement {Pλ}λ⊢d N, where Pλ is the
orthogonal projec on onto [λ]⊗ Vλ.
▶ Intui on: {Pλ}λ⊢d N respects the Ud- and SN-symmetries of the
spectrum es ma on problem.
Spectrum es ma on [Keyl and Werner 2001]
Let YN denote the outcome of the measurement {Pλ}λ⊢d N applied to
ρ⊗N where ρ is a state. Then, as N → ∞,
1NYN
D−−→ spec(ρ).
26 / 35
Spectrum es ma on & random matrix theory
▶ For the completely mixed state τ = 1d1, the corresponding
probability distribu on is called Schur-Weyl distribu on:
pd,N(λ) = Tr(Pλτ⊗N) =1dN
dλmd,λ.
▶ Spectrum es ma on: For the RV YτN obtained from applying the
measurement {Pλ}λ⊢d N to τ, we have
1NY
τN
D−−→ (1/d, . . . , 1/d).
▶ What about a ”central limit theorem” version of this describing
fluctua ons of Young diagrams?
27 / 35
Spectrum es ma on & random matrix theory
▶ To make this exact, define the centered and normalized RV
AN =λN − (N/d, . . . ,N/d)√
N/d
where λN ∼ pd,N takes values in Young diagrams {λ `d N}.
▶ LetM be the Gaussian unitary ensemble GUE(d): a Hermi an
random matrix whose entries are independent Gaussian RVs.
(df: exp(− 12 TrH
2) where H is a Hermi an matrix-valued RV.)
▶ DefineM0 = M− Tr(M)d 1, called the traceless Gaussian unitary
ensemble GUE0(d).
28 / 35
Main technical result
Fluctua ons of Schur-Weyl distribu on [Johansson 2001]
For the RV AN =√
dN(λN − (N/d, . . . ,N/d)),
AND−−→ spec(G),
where G ∼ GUE0(d).
Note that spec(G) d→∞−−−→a.s. Wigner’s semicircle law.
Main technical result (informal)
Strengthening of Johansson’s result:
E[g(AN)]n→∞−−−→ E[g(spec(G))]
for ”suitable” func ons g.
29 / 35
Applica on of Johansson strengthening
▶ Proof idea of asympto cs for standard and EPR protocol:
apply convergence of expecta on values to exact RT formulas by
rewri ng them as expecta on values over Schur-Weyl
distribu on.
▶ Main principle: Compu ng expecta on values of (func ons of)
GUE-distributed matrices is much easier!
▶ Example: probabilis c PBT
pEPRd =1dN
∑α ⊢d N−1
m2αdμ∗mμ∗
▶ Rewrite pEPRd =NdEα[(α1 + d)−1] and apply technical result.
30 / 35
More results in the paper
Main result 3: fully op mized determinis c PBT
▶ Achievability bound:
F∗d ≥ 1− d5 + O(d9/2)
4√2N2
+ O(N−3).
▶ Converse bound:
F∗d ≤ 1− d2 − 116N2 .
▶ Asympto cs of op mal determinis c PBT are given by
F∗d = 1− Θ(N−2).
31 / 35
Table of Contents
1 Opera onal se ng & known bounds
2 Symmetries & representa on theory
3 Main results: Asympto cs of PBT protocols
4 Proof methods
5 Concluding remarks
32 / 35
Summary
▶ We discussed two variants of port-based teleporta on (PBT):
▷ determinis c PBT with entanglement fidelity Fd;
▷ probabilis c PBT with success probability pd.
▶ Inherent symmetries: closed representa on-theore c formulas
for Fd and pd. [Studziński et al. 2017; Mozrzymas et al. 2017]
▶ Standard protocols: use connec on between Young diagrams
and GUE to determine asympto cs.
▶ We also determine asympto cs of fully op mized case using
different proof technique.
33 / 35
Open problems
Connec on between Schur-Weyl distribu on and GUE seems very
frui ul: get asympto cs for other ”symmetric” tasks?
For the op mal determinis c case, achievability bound is op mal in
N, but not in d-dependence: improvement?
We considered natural limit of fixed d and N → ∞.
What about the limit N, d → ∞ with N/d2 = const?
34 / 35
References
Beigi, S. and R. König (2011). New Journal of Physics 13.9, p. 093036. arXiv: 1101.1065 [quant-ph].Benne , C. H. et al. (1993). Physical Review Le ers 70.13, p. 1895.
Buhrman, H. et al. (2014). SIAM Journal on Compu ng 43.1, pp. 150–178. arXiv: 1009.2490 [quant-ph].Ishizaka, S. (2015). arXiv preprint. arXiv: 1506.01555 [quant-ph].Ishizaka, S. and T. Hiroshima (2008). Physical Review Le ers 101.24, p. 240501. arXiv: 0807.4568 [quant-ph].— (2009). Physical Review A 79.4, p. 042306. arXiv: 0901.2975 [quant-ph].Johansson, K. (2001). Annals of Mathema cs 153.1, pp. 259–296. arXiv: math/9906120 [math.CO].Keyl, M. and R. F. Werner (2001). Physical Review A 64.5, p. 052311. arXiv: quant-ph/0102027.Mozrzymas, M. et al. (2017). New Journal of Physics 20.5, p. 053006. arXiv: 1707.08456 [quant-ph].Nielsen, M. A. and I. L. Chuang (1997). Physical Review Le ers 79.2, p. 321. arXiv: quant-ph/9703032.Pirandola, S. et al. (2018). npj Quantum Informa on 5, p. 50. arXiv: 1803.02834 [quant-ph].Studziński, M. et al. (2017). Scien fic Reports 7, p. 10871. arXiv: 1612.09260 [quant-ph].
Thank you very much for your a en on!
35 / 35