Distinguishing multi-partite states by local measurements
Guillaume Aubrun / Andreas Winter
Université Claude Bernard Lyon 1 / Universitat Autònoma de Barcelona
Cécilia Lancien
UCBL1 / UAB
Outline
1 Introduction
2 Distinguishability norm associated with a single local measurement on a multi-partite system
3 Distinguishability norm associated with whole classes of locally restricted measurements on a multi-partite system
4 Conclusion and open questions
Outline
1 Introduction
2 Distinguishability norm associated with a single local measurement on a multi-partite system
3 Distinguishability norm associated with whole classes of locally restricted measurements on a multi-partite system
4 Conclusion and open questions
Distinguishability norms (1)
Situation considered
System that can be in 2 quantum states,ρorσ, with respective prior probabilitiesqand 1
−
q.→
Decide in which one it is most likely, based on accessible experimental data, i.e. on the outcomes of a POVMM= (
Mx)
x∈X performed on it.Probability of error :PE
=
12 1
− ∑
x∈X
Tr
qρ
− (
1−
q)σ
Mx
!
Distinguishability (semi)-norm associated with the POVMM
= (
Mx)
x∈X :k∆k
M:= ∑
x∈X
Tr
∆
MxSetMof POVMs :
k∆k
M:=
supM∈M
k∆k
M→
Holevo-Helstrom :k∆k
ALL= k∆k
1:=
Tr∆
Distinguishability norms (1)
Situation considered
System that can be in 2 quantum states,ρorσ, with respective prior probabilitiesqand 1
−
q.→
Decide in which one it is most likely, based on accessible experimental data, i.e. on the outcomes of a POVMM= (
Mx)
x∈X performed on it.Probability of error :PE
=
12 1
− ∑
x∈X
Tr
qρ
− (
1−
q)σ
Mx
!
Distinguishability (semi)-norm associated with the POVMM
= (
Mx)
x∈X :k∆k
M:= ∑
x∈X
Tr
∆
MxSetMof POVMs :
k∆k
M:=
supM∈M
k∆k
M→
Holevo-Helstrom :k∆k
ALL= k∆k
1:=
Tr∆
Distinguishability norms (1)
Situation considered
System that can be in 2 quantum states,ρorσ, with respective prior probabilitiesqand 1
−
q.→
Decide in which one it is most likely, based on accessible experimental data, i.e. on the outcomes of a POVMM= (
Mx)
x∈X performed on it.Probability of error :PE
=
12 1
− ∑
x∈X
Tr
qρ
− (
1−
q)σ
Mx
!
Distinguishability (semi)-norm associated with the POVMM
= (
Mx)
x∈X :k∆k
M:= ∑
x∈X
Tr
∆
MxSetMof POVMs :
k∆k
M:=
supM∈M
k∆k
M→
Holevo-Helstrom :k∆k
ALL= k∆k
1:=
Tr∆
Distinguishability norms (1)
Situation considered
System that can be in 2 quantum states,ρorσ, with respective prior probabilitiesqand 1
−
q.→
Decide in which one it is most likely, based on accessible experimental data, i.e. on the outcomes of a POVMM= (
Mx)
x∈X performed on it.Probability of error :PE
=
12 1
− ∑
x∈X
Tr
qρ
− (
1−
q)σ
Mx
!
Distinguishability (semi)-norm associated with the POVMM
= (
Mx)
x∈X :k∆k
M:= ∑
x∈X
Tr
∆
MxDistinguishability norms (2)
Problem
When only POVMs from a restricted setMare allowed, how much smaller than
k · k
1isk · k
M?What kind of “restrictions” ? On a multi-partite system, experimenters are not able to implement any observable (measurements on their own sub-system).
→
Mis often defined by these locality constraints (e.g.LOCC⊂
SEP⊂
PPT).Motivation :Existence of “Data-Hiding” states(DiVincenzo/Leung/Terhal) Orthogonal states (hence perfectly distinguishable by a suitable measurement) that are barely distinguishable by PPT (and even more so LOCC)
measurements.
→
Ex in the bipartite case :Completely symmetric and antisymmetric states onCD⊗
CD,σ:=
D21+D(1 +
F)andα:=
D21−D(1 −
F).∆ :=
1 2σ−
12αis s.t.
k∆k
PPT=
2D
+
1 1= k∆k
1.Distinguishability norms (2)
Problem
When only POVMs from a restricted setMare allowed, how much smaller than
k · k
1isk · k
M?What kind of “restrictions” ? On a multi-partite system, experimenters are not able to implement any observable (measurements on their own sub-system).
→
Mis often defined by these locality constraints (e.g.LOCC⊂
SEP⊂
PPT).Motivation :Existence of “Data-Hiding” states(DiVincenzo/Leung/Terhal) Orthogonal states (hence perfectly distinguishable by a suitable measurement) that are barely distinguishable by PPT (and even more so LOCC)
measurements.
→
Ex in the bipartite case :Completely symmetric and antisymmetric statesD
⊗
D 1 1−
Outline
1 Introduction
2 Distinguishability norm associated with a single local measurement on a multi-partite system
3 Distinguishability norm associated with whole classes of locally restricted measurements on a multi-partite system
4 Conclusion and open questions
Local measurements on a multi-partite quantum system
Finite-dimensional multi-partite quantum system :
H =
Cd1⊗ · · · ⊗
CdK withd:=
dimH =
d1× · · · ×
dK <+∞
M
=
M(1)⊗ · · · ⊗
M(K)a local POVM onH
. Comparison ofk · k
Mwith :k · k
1→
most natural ?k · k
2→
most relevant ! (norm equivalence with dimension-independent constants of domination)Definition
K-partite generalization of the 2-norm :
k∆k
2(K):=
s∑
I⊂[K]
Tr TrI
∆
2
Remark :On a single system,
k∆k
2(1)=
p|
Tr∆|
2+
Tr|∆|
2reduces tok∆k
2=
pTr
|∆|
2on traceless Hermitians∆
.Local measurements on a multi-partite quantum system
Finite-dimensional multi-partite quantum system :
H =
Cd1⊗ · · · ⊗
CdK withd:=
dimH =
d1× · · · ×
dK <+∞
M
=
M(1)⊗ · · · ⊗
M(K)a local POVM onH
. Comparison ofk · k
Mwith :k · k
1→
most natural ?k · k
2→
most relevant ! (norm equivalence with dimension-independent constants of domination)Definition
K-partite generalization of the 2-norm :
k∆k
2(K):=
s∑
I⊂[K]
Tr TrI
∆
2
Remark :On a single system,
k∆k
2(1)=
p|
Tr∆|
2+
Tr|∆|
2reduces tok∆k
2=
pTr
|∆|
2on traceless Hermitians∆
.Local measurements on a multi-partite quantum system
Finite-dimensional multi-partite quantum system :
H =
Cd1⊗ · · · ⊗
CdK withd:=
dimH =
d1× · · · ×
dK <+∞
M
=
M(1)⊗ · · · ⊗
M(K)a local POVM onH
. Comparison ofk · k
Mwith :k · k
1→
most natural ?k · k
2→
most relevant ! (norm equivalence with dimension-independent constants of domination)Definition
K-partite generalization of the 2-norm :
k∆k
2(K):=
s
∑
I⊂[K]
Tr TrI
∆
2
Remark :On a single system,
k∆k
2(1)=
p|
Tr∆|
2+
Tr|∆|
2reduces tok∆k
2=
pTr
|∆|
2on traceless Hermitians∆
.Local measurements on a multi-partite quantum system
Finite-dimensional multi-partite quantum system :
H =
Cd1⊗ · · · ⊗
CdK withd:=
dimH =
d1× · · · ×
dK <+∞
M
=
M(1)⊗ · · · ⊗
M(K)a local POVM onH
. Comparison ofk · k
Mwith :k · k
1→
most natural ?k · k
2→
most relevant ! (norm equivalence with dimension-independent constants of domination)Definition
K-partite generalization of the 2-norm :
k∆k
2(K):=
s
∑
I⊂[K]
Tr TrI
∆
2
Remark :On a single system,
k∆k
2(1)=
p|
Tr∆|
2+
Tr|∆|
2reduces tok∆k
2=
pTr
|∆|
2on traceless Hermitians∆
.Tensor product of local 4-design POVMs (1)
M
:=
DpxPxx∈X is at-design POVM onCDif
(
px)
x∈X is a probability distribution and(
Px)
x∈X are rank-1 projectors onCDs.t.∑
x∈X
pxPx⊗t
=
Z|ψi∈CD,hψ|ψi=1
|ψihψ|
⊗tdψ=
1D
× · · · × (
D+
t−
1) ∑
σ∈St
Uσ.
Example :U
:= {
D|ψihψ|dψ, |ψi ∈
CD,hψ|ψi =
1}
∞-design POVM onCD.Lemma :
“Moments’ method”M
=
M(1)⊗ · · · ⊗
M(K)withM(j):=
djpxjPxj
xj∈Xj 4-design POVM onCdj.
∆
an Hermitian onH
, andSthe random variable taking value Tr∆
Px1⊗ · · · ⊗
PxKwith probabilitypx1
× · · · ×
pxK, so thatk∆k
M=
d E|
S|
. By Jensen :E|
S| ≤
pE
(
S2)
, and by Hölder :E|
S| ≥
q(E(S2))3 E(S4) . So :d
s
(
E(
S2))
3E(S4
) ≤ k∆k
M≤
dp E(
S2)
.Tensor product of local 4-design POVMs (1)
M
:=
DpxPxx∈X is at-design POVM onCDif
(
px)
x∈X is a probability distribution and(
Px)
x∈X are rank-1 projectors onCDs.t.∑
x∈X
pxPx⊗t
=
Z|ψi∈CD,hψ|ψi=1
|ψihψ|
⊗tdψ=
1D
× · · · × (
D+
t−
1) ∑
σ∈St
Uσ.
Example :U
:= {
D|ψihψ|dψ, |ψi ∈
CD,hψ|ψi =
1}
∞-design POVM onCD.Lemma :
“Moments’ method”M
=
M(1)⊗ · · · ⊗
M(K)withM(j):=
djpxjPxj
xj∈Xj 4-design POVM onCdj.
∆
an Hermitian onH
, andSthe random variable taking value Tr∆
Px1⊗ · · · ⊗
PxKwith probabilitypx1
× · · · ×
pxK, so thatk∆k
M=
d E|
S|
. By Jensen :E|
S| ≤
pE
(
S2)
, and by Hölder :E|
S| ≥
q(E(S2))3 E(S4) . So :d
s
(
E(
S2))
3E(S4
) ≤ k∆k
M≤
dp E(
S2)
.Tensor product of local 4-design POVMs (1)
M
:=
DpxPxx∈X is at-design POVM onCDif
(
px)
x∈X is a probability distribution and(
Px)
x∈X are rank-1 projectors onCDs.t.∑
x∈X
pxPx⊗t
=
Z|ψi∈CD,hψ|ψi=1
|ψihψ|
⊗tdψ=
1D
× · · · × (
D+
t−
1) ∑
σ∈St
Uσ.
Example :U
:= {
D|ψihψ|dψ, |ψi ∈
CD,hψ|ψi =
1}
∞-design POVM onCD.Lemma :
“Moments’ method”M
=
M(1)⊗ · · · ⊗
M(K)withM(j):=
djpxjPxj
xj∈Xj 4-design POVM onCdj.
∆
an Hermitian onH
, andSthe random variable taking value Tr∆
Px1⊗ · · · ⊗
PxKwith probabilitypx1
× · · · ×
pxK, so thatk∆k
M=
d E|
S|
.By Jensen :E
|
S| ≤
pE
(
S2)
, and by Hölder :E|
S| ≥
q(E(S2))3 E(S4) . So :d
s
(
E(
S2))
3E(S4
) ≤ k∆k
M≤
dp E(
S2)
.Tensor product of local 4-design POVMs (1)
M
:=
DpxPxx∈X is at-design POVM onCDif
(
px)
x∈X is a probability distribution and(
Px)
x∈X are rank-1 projectors onCDs.t.∑
x∈X
pxPx⊗t
=
Z|ψi∈CD,hψ|ψi=1
|ψihψ|
⊗tdψ=
1D
× · · · × (
D+
t−
1) ∑
σ∈St
Uσ.
Example :U
:= {
D|ψihψ|dψ, |ψi ∈
CD,hψ|ψi =
1}
∞-design POVM onCD.Lemma :
“Moments’ method”M
=
M(1)⊗ · · · ⊗
M(K)withM(j):=
djpxjPxj
xj∈Xj 4-design POVM onCdj.
∆
an Hermitian onH
, andSthe random variable taking value Tr∆
Px1⊗ · · · ⊗
PxKwith probabilitypx1
× · · · ×
pxK, so thatk∆k
M=
d E|
S|
. By Jensen :E|
S| ≤
pE
(
S2)
, and by Hölder :E|
S| ≥
q(E(S2))3 E(S4) .
So :d s
(
E(
S2))
3E(S4
) ≤ k∆k
M≤
dp E(
S2)
.Tensor product of local 4-design POVMs (1)
M
:=
DpxPxx∈X is at-design POVM onCDif
(
px)
x∈X is a probability distribution and(
Px)
x∈X are rank-1 projectors onCDs.t.∑
x∈X
pxPx⊗t
=
Z|ψi∈CD,hψ|ψi=1
|ψihψ|
⊗tdψ=
1D
× · · · × (
D+
t−
1) ∑
σ∈St
Uσ.
Example :U
:= {
D|ψihψ|dψ, |ψi ∈
CD,hψ|ψi =
1}
∞-design POVM onCD.Lemma :
“Moments’ method”M
=
M(1)⊗ · · · ⊗
M(K)withM(j):=
djpxjPxj
xj∈Xj 4-design POVM onCdj.
∆
an Hermitian onH
, andSthe random variable taking value Tr∆
Px1⊗ · · · ⊗
PxKwith probabilitypx1
× · · · ×
pxK, so thatk∆k
M=
d E|
S|
. By Jensen :E|
S| ≤
pE
(
S2)
, and by Hölder :E|
S| ≥
q(E(S2))3 E(S4) .
Tensor product of local 4-design POVMs (2)
Theorem
(L/Winter)M
=
M(1)⊗ · · · ⊗
M(K)withM(j)4-design POVM onCdj. E(S2) =
K
∏
j=1
1
dj
(
dj+
1) ∑
I⊂[K]
Tr TrI
∆
2E
(
S4) ≤
K
∏
j=1
1
dj
× · · · × (
dj+
3)
18K
"
∑
I⊂[K]
Tr TrI
∆
2#2
So : 1
18K/2
k∆k
2(K)≤ k∆k
M≤ k∆k
2(K).Conclusion :IfM is a “sufficiently symmetric” local POVM on
H
(tensorproduct of local 4-design POVMs), then
k · k
Mis “essentially” equivalent tok · k
2(K)(dimension-independent constants of domination).Tensor product of local 4-design POVMs (2)
Theorem
(L/Winter)M
=
M(1)⊗ · · · ⊗
M(K)withM(j)4-design POVM onCdj. E(S2) =
K
∏
j=1
1
dj
(
dj+
1) ∑
I⊂[K]
Tr TrI
∆
2E
(
S4) ≤
K
∏
j=1
1
dj
× · · · × (
dj+
3)
18K
"
∑
I⊂[K]
Tr TrI
∆
2#2
So : 1
18K/2
k∆k
2(K)≤ k∆k
M≤ k∆k
2(K).Conclusion :IfM is a “sufficiently symmetric” local POVM on
H
(tensorproduct of local 4-design POVMs), then
k · k
Mis “essentially” equivalent tok · k
2(K)(dimension-independent constants of domination).Tensor product of local 4-design POVMs (2)
Theorem
(L/Winter)M
=
M(1)⊗ · · · ⊗
M(K)withM(j)4-design POVM onCdj. E(S2) =
K
∏
j=1
1
dj
(
dj+
1) ∑
I⊂[K]
Tr TrI
∆
2E
(
S4) ≤
K
∏
j=1
1
dj
× · · · × (
dj+
3)
18K
"
∑
I⊂[K]
Tr TrI
∆
2#2
So : 1
18K/2
k∆k
2(K)≤ k∆k
M≤ k∆k
2(K).Conclusion :IfM is a “sufficiently symmetric” local POVM on
H
(tensorproduct of local 4-design POVMs), then
k · k
Mis “essentially” equivalent tok · k
2(K)(dimension-independent constants of domination).Related work and applications
Remark :Other 4thvs2ndorder moment inequalities may be obtained from hypercontractive inequalities(Montanaro).
Previously known results :Let
∆
be a traceless Hermitian onH
.K
=
1: 13k∆k
2≤ k∆k
M≤ k∆k
2forM a 4-design POVM.→
Applications in quantum algorithms(Ambainis/Emerson)and quantum dimensionality reduction(Harrow/Montanaro/Short).K
=
2: √1153
k∆k
2≤ k∆k
MforM a tensor product of two 4-design POVMs(Matthews/Wehner/Winter).→
Applications in entanglement theory(Brandão/Christandl/Yard).→
Results for anyK and non-necessarily traceless∆
useful too ?Related work and applications
Remark :Other 4thvs2ndorder moment inequalities may be obtained from hypercontractive inequalities(Montanaro).
Previously known results :Let
∆
be a traceless Hermitian onH
.K
=
1: 13k∆k
2≤ k∆k
M≤ k∆k
2forM a 4-design POVM.→
Applications in quantum algorithms(Ambainis/Emerson)and quantum dimensionality reduction(Harrow/Montanaro/Short).K
=
2: √1153
k∆k
2≤ k∆k
MforM a tensor product of two 4-design POVMs(Matthews/Wehner/Winter).→
Applications in entanglement theory(Brandão/Christandl/Yard).→
Results for anyK and non-necessarily traceless∆
useful too ?“Implementable” POVM with “good” discriminating power (1)
Problem
A 4-design POVM onCd must have at least
Ω(
d4)
outcomes.No explicit constructions of 4-design POVMs are known.
→
Minimal number of outcomes for a randomly chosen POVMMonCdso that with high probabilityk · k
M' k · k
2(1)(dimension-independently) ?Strategy :Draw independentlyP1, . . . ,Pnuniformly distributed rank-1 projectors onCd, setS
:=
∑nk=1
Pk andP
:= (e
Pk:=
S−1/2PkS−1/2)
1≤k≤n. Pis a random POVM onCdwithnoutcomes s.t.k∆k
P=
n
∑
k=1
Tr(∆ePk
)
.Theorem
(Aubrun/L)∀α
>0,∃
Cα>0:
n≥
Cαd2⇒
P 18√
18
k · k
2(1)≤ k · k
P≤
158k · k
2(1)≥
1−
αRemark :“Optimal result” sincePinformationally complete
⇒
n≥
d2.“Implementable” POVM with “good” discriminating power (1)
Problem
A 4-design POVM onCd must have at least
Ω(
d4)
outcomes.No explicit constructions of 4-design POVMs are known.
→
Minimal number of outcomes for a randomly chosen POVMMonCdso that with high probabilityk · k
M' k · k
2(1)(dimension-independently) ?Strategy :Draw independentlyP1, . . . ,Pnuniformly distributed rank-1 projectors onCd, setS
:=
∑nk=1
Pk andP
:= (e
Pk:=
S−1/2PkS−1/2)
1≤k≤n. Pis a random POVM onCdwithnoutcomes s.t.k∆k
P=
n
∑
k=1
Tr(∆ePk
)
.Theorem
(Aubrun/L)∀α
>0,∃
Cα>0:
n≥
Cαd2⇒
P 18√
18
k · k
2(1)≤ k · k
P≤
158k · k
2(1)≥
1−
αRemark :“Optimal result” sincePinformationally complete
⇒
n≥
d2.“Implementable” POVM with “good” discriminating power (1)
Problem
A 4-design POVM onCd must have at least
Ω(
d4)
outcomes.No explicit constructions of 4-design POVMs are known.
→
Minimal number of outcomes for a randomly chosen POVMMonCdso that with high probabilityk · k
M' k · k
2(1)(dimension-independently) ?Strategy :Draw independentlyP1, . . . ,Pnuniformly distributed rank-1 projectors onCd, setS
:=
∑nk=1
Pk andP
:= (e
Pk:=
S−1/2PkS−1/2)
1≤k≤n. Pis a random POVM onCdwithnoutcomes s.t.k∆k
P=
n
∑
k=1
Tr(∆ePk
)
.Theorem
(Aubrun/L)
Remark :“Optimal result” sincePinformationally complete
⇒
n≥
d2.“Implementable” POVM with “good” discriminating power (1)
Problem
A 4-design POVM onCd must have at least
Ω(
d4)
outcomes.No explicit constructions of 4-design POVMs are known.
→
Minimal number of outcomes for a randomly chosen POVMMonCdso that with high probabilityk · k
M' k · k
2(1)(dimension-independently) ?Strategy :Draw independentlyP1, . . . ,Pnuniformly distributed rank-1 projectors onCd, setS
:=
∑nk=1
Pk andP
:= (e
Pk:=
S−1/2PkS−1/2)
1≤k≤n. Pis a random POVM onCdwithnoutcomes s.t.k∆k
P=
n
∑
k=1
Tr(∆ePk
)
.Theorem
(Aubrun/L)∀α
>0,∃
Cα>0:
n≥
Cαd2⇒
P 18√
18
k · k
2(1)≤ k · k
P≤
158k · k
2(1)≥
1−
αRemark :“Optimal result” sincePinformationally complete
⇒
n≥
d2.“Implementable” POVM with “good” discriminating power (2)
Idea of the proof :
1 Large deviation probability of dn
n
∑
k=1
|Tr(
Pk∆)|
fromk∆k
2(1)?2 Large deviation probability of
n
∑
k=1
|Tr(e
Pk∆)|
from dnn
∑
k=1
|Tr(
Pk∆)|
?→
Use twice :(i) Moments’ estimate :
∀
p≥
1,E|
X|
p≤ (
ecp)
p⇒ k
Xk
ψ1 .ec (ii) Bernstein-type tail bound for sums of i.i.d centeredψ1-r.v. (iii) Net argument : individual to global error term1
{
Xk:=
dTr Pk
∆
− k∆k
U}
1≤k≤nfor a given∆ ∈
S2(1)2
{
Xk:= hψ|
dPk−
1|ψi}1≤k≤nfor a given unit vector|ψi
P1 8
√
18
k · k
2(1)≤ k · k
P≤
158
k · k
2(1)≥
1−
Cd2e−cn−
C0de−c0n/d“Implementable” POVM with “good” discriminating power (2)
Idea of the proof :
1 Large deviation probability of dn
n
∑
k=1
|Tr(
Pk∆)|
fromk∆k
2(1)?2 Large deviation probability of
n
∑
k=1
|Tr(e
Pk∆)|
from dnn
∑
k=1
|Tr(
Pk∆)|
?→
Use twice :(i) Moments’ estimate :
∀
p≥
1,E|
X|
p≤ (
ecp)
p⇒ k
Xk
ψ1 .ec (ii) Bernstein-type tail bound for sums of i.i.d centeredψ1-r.v.(iii) Net argument : individual to global error term
1
{
Xk:=
dTr Pk
∆
− k∆k
U}
1≤k≤nfor a given∆ ∈
S2(1)2
{
Xk:= hψ|
dPk−
1|ψi}1≤k≤nfor a given unit vector|ψi
P1 8
√
18
k · k
2(1)≤ k · k
P≤
158
k · k
2(1)≥
1−
Cd2e−cn−
C0de−c0n/d“Implementable” POVM with “good” discriminating power (2)
Idea of the proof :
1 Large deviation probability of dn
n
∑
k=1
|Tr(
Pk∆)|
fromk∆k
2(1)?2 Large deviation probability of
n
∑
k=1
|Tr(e
Pk∆)|
from dnn
∑
k=1
|Tr(
Pk∆)|
?→
Use twice :(i) Moments’ estimate :
∀
p≥
1,E|
X|
p≤ (
ecp)
p⇒ k
Xk
ψ1 .ec (ii) Bernstein-type tail bound for sums of i.i.d centeredψ1-r.v.(iii) Net argument : individual to global error term
1
{
Xk:=
dTr Pk
∆
− k∆k
U}
1≤k≤nfor a given∆ ∈
S2(1)2
{
Xk:= hψ|
dPk−
1|ψi}1≤k≤nfor a given unit vector|ψi
P 1
8
√
18
k · k
2(1)≤ k · k
P≤
158
k · k
2(1)≥
1−
Cd2e−cn−
C0de−c0n/d“Implementable” POVM with “good” discriminating power (2)
Idea of the proof :
1 Large deviation probability of dn
n
∑
k=1
|Tr(
Pk∆)|
fromk∆k
2(1)?2 Large deviation probability of
n
∑
k=1
|Tr(e
Pk∆)|
from dnn
∑
k=1
|Tr(
Pk∆)|
?→
Use twice :(i) Moments’ estimate :
∀
p≥
1,E|
X|
p≤ (
ecp)
p⇒ k
Xk
ψ1 .ec (ii) Bernstein-type tail bound for sums of i.i.d centeredψ1-r.v.(iii) Net argument : individual to global error term
1
{
Xk:=
dTr Pk
∆
− k∆k
U}
1≤k≤nfor a given∆ ∈
S2(1)2
{
Xk:= hψ|
dPk−
1|ψi}1≤k≤nfor a given unit vector|ψi
P1 8
√
18
k · k
2(1)≤ k · k
P≤
158
k · k
2(1)≥
1−
Cd2e−cn−
C0de−c0n/dOutline
1 Introduction
2 Distinguishability norm associated with a single local measurement on a multi-partite system
3 Distinguishability norm associated with whole classes of locally restricted measurements on a multi-partite system
4 Conclusion and open questions
Bounds on the distinguishability of multi-partite states under LOCC, SEP and PPT measurements
A tensor product of 4-design POVMs is a particular LOCC strategy. Consequently : 1
18K/2
k · k
2≤
118K/2
k · k
2(K)≤ k · k
LOCC.On am-partite Hilbert space, the ball of radius 21−m/2for
k · k
2(centered at1) is fully separable(Barnum/Gurvits).Consequently : 2
2K/2
k · k
2≤ k · k
SEPandk · k
2≤ k · k
PPT.Comparison of k · k
LOCC, k · k
SEPand k · k
PPTwith k · k
ALL1 18K/2
√
1d
k · k
1≤ k · k
LOCC≤ k · k
1 22K/2
√
1d
k · k
1≤ k · k
SEP≤ k · k
1√
1d
k · k
1≤ k · k
PPT≤ k · k
1Bounds on the distinguishability of multi-partite states under LOCC, SEP and PPT measurements
A tensor product of 4-design POVMs is a particular LOCC strategy.
Consequently : 1
18K/2
k · k
2≤
118K/2
k · k
2(K)≤ k · k
LOCC.On am-partite Hilbert space, the ball of radius 21−m/2for
k · k
2(centered at1) is fully separable(Barnum/Gurvits).Consequently : 2
2K/2
k · k
2≤ k · k
SEPandk · k
2≤ k · k
PPT.Comparison of k · k
LOCC, k · k
SEPand k · k
PPTwith k · k
ALL1 18K/2
√
1d
k · k
1≤ k · k
LOCC≤ k · k
1 22K/2
√
1d
k · k
1≤ k · k
SEP≤ k · k
1√
1d
k · k
1≤ k · k
PPT≤ k · k
1Bounds on the distinguishability of multi-partite states under LOCC, SEP and PPT measurements
A tensor product of 4-design POVMs is a particular LOCC strategy.
Consequently : 1
18K/2
k · k
2≤
118K/2
k · k
2(K)≤ k · k
LOCC.On am-partite Hilbert space, the ball of radius 21−m/2for
k · k
2(centered at1) is fully separable(Barnum/Gurvits).Consequently : 2
2K/2
k · k
2≤ k · k
SEPandk · k
2≤ k · k
PPT.Comparison of k · k
LOCC, k · k
SEPand k · k
PPTwith k · k
ALL 118K/2
√
1d
k · k
1≤ k · k
LOCC≤ k · k
1 22K/2
√
1d
k · k
1≤ k · k
SEP≤ k · k
1√
1d
k · k
1≤ k · k
PPT≤ k · k
1Bounds on the distinguishability of multi-partite states under LOCC, SEP and PPT measurements
A tensor product of 4-design POVMs is a particular LOCC strategy.
Consequently : 1
18K/2
k · k
2≤
118K/2
k · k
2(K)≤ k · k
LOCC.On am-partite Hilbert space, the ball of radius 21−m/2for
k · k
2(centered at1) is fully separable(Barnum/Gurvits).Consequently : 2
2K/2
k · k
2≤ k · k
SEPandk · k
2≤ k · k
PPT.Comparison of k · k
LOCC, k · k
SEPand k · k
PPTwith k · k
ALL1 18K/2
√
1d
k · k
1≤ k · k
LOCC≤ k · k
12 1
√ k · k ≤ k · k ≤ k · k
Data-Hiding and optimality of the lower bounds
M
:=
UCd1⊗ · · · ⊗
UCdK tensor product of the local uniform POVMs. Tightness of 118K/2
√1
d
k · k
1≤ k · k
M?∃ ∆ 6=
0: k∆k
M≤
δ√K/2d
k∆k
1with 2π<δ<1
→
Dependence onK anddof the constant relating the norms is “real”. Tightness of √1d
k · k
1≤ k · k
PPT? If∃
I⊂ [
K] :
∏i∈I
di
=
∏i∈/I
di
= √
d, then
∃ ∆ 6=
0: k∆k
PPT≤
√2d+1
k∆k
1→
Dependence ond of the constant relating the norms is “real”. Remark :In the special caseH = (
CD)
⊗K(
d=
DK)
withK fixed and D→ +∞
, it may be shown by volumic considerations that “typically” : (k · k
SEP'
√ 1d1−1/K
k · k
1k · k
PPT' k · k
1 , so that :k · k
PPTk · k
SEP √1d
k · k
1(Aubrun/L).→
Data-Hiding Hermitians are “exceptionnal”.Data-Hiding and optimality of the lower bounds
M
:=
UCd1⊗ · · · ⊗
UCdK tensor product of the local uniform POVMs.Tightness of 1
18K/2
√1
d
k · k
1≤ k · k
M?∃ ∆ 6=
0: k∆k
M≤
δ√K/2d
k∆k
1with 2π<δ<1
→
Dependence onK anddof the constant relating the norms is “real”.Tightness of √1
d
k · k
1≤ k · k
PPT? If∃
I⊂ [
K] :
∏i∈I
di
=
∏i∈/I
di
= √
d, then
∃ ∆ 6=
0: k∆k
PPT≤
√2d+1
k∆k
1→
Dependence ond of the constant relating the norms is “real”. Remark :In the special caseH = (
CD)
⊗K(
d=
DK)
withK fixed and D→ +∞
, it may be shown by volumic considerations that “typically” : (k · k
SEP'
√ 1d1−1/K
k · k
1k · k
PPT' k · k
1 , so that :k · k
PPTk · k
SEP √1d