Error Exponent Analysis in Quantum Source and Channel Coding
1. [arXiv:1803.07505]: Cheng (NTU&UTS), Hanson (Cambridge), Datta (Cambridge), MH
3. [arXiv:1701.03195]: Cheng (NTU&UTS), MH
2. [arXiv:1704.05703]: Cheng (NTU&UTS), MH, Tomamichel (UTS)
In Source/Channel Coding,
- Error Probability:
- Transmission Rate:
- Code Length:
ε
\varepsilon
R
R
n
n
ε
\varepsilon
R
R
C
C
1
1
n=1
n=1
n=∞
n=\infty
Small Deviation
R→C
R\to C
ε>0
\varepsilon>0
Large Deviation
ε→0
\varepsilon\to 0
R=C
R \neq C
Channel Coding Trade-offs
Moderate Deviation
R→C
R\to C
ε→0
\varepsilon\to 0
ε
\varepsilon
R
R
C
C
1
1
n=1
n=1
n=∞
n=\infty
Small Deviation
R→C
R\to C
ε>0
\varepsilon>0
Large Deviation
ε→0
\varepsilon\to 0
R=C
R \neq C
Source Coding Trade-offs
Moderate Deviation
R→C
R\to C
ε→0
\varepsilon\to 0
Three Regimes:
Small Deviation
Large Deviation
Moderate Deviation
R→C
R\to C
ε=0
\varepsilon \neq0
ε→0
\varepsilon\to 0
R=C
R \neq C
ε→0
\varepsilon\to 0
R→C
R\to C
Small Deviation:
a.k.a. Second-Order Analysis
Rn(ε)=CC+nVCΦ−1(ε)+O(nlogn)
R_n(\varepsilon) = C_C + \sqrt{\frac{V_C}{n}}\Phi^{-1}(\varepsilon) + O(\frac{\log n}{n})
Strassen, Transactions of the Third Prague Conference on Information Theory, pp. 689–723, 1962.
Tomamichel and Tan, CMP 338(1):103–137, 2015.
Rn(ε)=CS−nVSΦ−1(ε)+O(nlogn)
R_n(\varepsilon) = C_S - \sqrt{\frac{V_S}{n}}\Phi^{-1}(\varepsilon) + O(\frac{\log n}{n})
Tomamichel and Hayashi, IEEE IT 59(11):7693-7710, 2013.
Nomura and Han. IEEE IT 60(9):5553–5572, 2014.
{
\{
{
\{
Source
Channel
Large Deviation:
a.k.a. Error Exponent Analysis
εn(R)=e−Θ(n)
\varepsilon_n(R) = e^{-\Theta(n)}
Shannon, Bell System Technical Journal, 38(3):611–656, 1959.
Burnashev and Holevo, Problems of information transmission, 34(2):97–107, 1998.
Moderate Deviation:
εn(Rn)=e−Θ(nan2)
\varepsilon_n(R_n) = e^{-\Theta(na_n^2)}
Altug and Wagner. IEEE TIT 60(8):4417–4426, 2014.
Chubb, Tomamichel and Tan, arXiv: 1701.03114.
{an}:an→0, ann→∞.
\{a_n\}: a_n\to 0,\ a_n\sqrt{n} \to \infty.
Rn=C−an
R_n= C- a_n
Cheng and Hsieh, arXiv: 1701.03195.
Outlines
Part I
Channel Coding
Xn
X^n
M^n
\hat{M}_n
En
\mathcal{E}_n
Mn
M_n
εn(R)=Pr(Mn=M^n): ∣Mn∣=2nR
\varepsilon_n(R) = \text{Pr}(M^n\neq \hat{M}^n):\ |M_n| = 2^{nR}
W⊗n
{W}^{\otimes n}
Dn
\mathcal{D}_n
Bn
B^n
Classical-Quantum Channels
W:X→Cd×d
W: \mathcal{X} \to \mathbb{C}^{d\times d}
limn→∞n−1logεn(R)=E(R)
\lim_{n\to\infty} \frac{-1}{n} \log\varepsilon_n(R) = E(R)
Error Exponent Analysis
Erc(R)≤E(R)≤Esp(R)
E_{\text{rc}}(R) \leq E(R) \leq E_{\text{sp}}(R)
Shannon, Bell System Technical Journal, 38(3):611–656, 1959.
Claissical Sphere-Packing Bounds
Esp(R):=sups≥0{maxPE0(s,P)−sR}
E_{\text{sp}}(R):=\sup_{s\geq 0} \left\{ \max_P E_0(s,P) -sR\right\}
E~sp(R):=maxPminV{D(V∥W∣P):I(P,V)≤R}
\tilde{E}_{\text{sp}}(R):=\max_P\min_V\{D(V\|W|P):I(P,V)\leq R\}
Shannon, Gallager, and Berlekamp. Information and Control, 10(1):65–103, 1967.
Haroutunian. Problemy Peredachi Informatsii, 4(4):37–48, 1968, (in Russian).
Blahut. IEEE TIT, 20(4):405–417, 1974.
Esp(R)=E~sp(R)
{E}_{\text{sp}}(R)= \tilde{E}_{\text{sp}}(R)
Quantum Sphere-Packing Bounds
Esp(R):=sups≥0{maxPE0(s,P)−sR}
E_{\text{sp}}(R):=\sup_{s\geq 0} \left\{ \max_P E_0(s,P) -sR\right\}
E~sp(R):=maxPminV{D(V∥W∣P):I(P,V)≤R}
\tilde{E}_{\text{sp}}(R):=\max_P\min_V\{D(V\|W|P):I(P,V)\leq R\}
Dalai. IEEE TIT, 59(12):8027–8056, 2013.
Winter. PhD Thesis, Universitate Bielefeld, 1999.
Cheng, Hsieh, and Tomamichel. arXiv: 1704.05703
Esp(R)≤E~sp(R)
{E}_{\text{sp}}(R)\leq \tilde{E}_{\text{sp}}(R)
E0(s,P)=−logTr[(∑xP(x)Wx1/1+s)1+s]
E_0(s,P)=-\log \text{Tr} \left[\left(\sum_{x}P(x)W_x^{1/1+s}\right)^{1+s}\right]
D(V∥W∣P)=∑xP(x)D(Vx∥Wx)
D(V\|W|P) = \sum_x P(x) D(V_x\|W_x)
E~sp(R,P)=sup0<α≤1minσ{α1−α(Dα♭(W∥σ∣P)−R)}
\tilde{E}_{\text{sp}}(R,P)=\sup_{0<\alpha\leq 1}\min_\sigma\{\frac{1-\alpha}{\alpha}\left(D^\flat_\alpha(W\|\sigma|P)-R\right)\}
Esp(R,P)≤sup0<α≤1minσ{α1−α(Dα(W∥σ∣P)−R)}
{E}_{\text{sp}}(R,P)\leq\sup_{0<\alpha\leq 1}\min_\sigma\{\frac{1-\alpha}{\alpha}\left(D_\alpha(W\|\sigma|P)-R\right)\}
Theorem:
Esp(R)≤E~sp(R)
{E}_{\text{sp}}(R)\leq \tilde{E}_{\text{sp}}(R)
Proof:
Dα(ρ∥σ):=α−11logTrρασ(1−α)≤Dα♭(ρ∥σ):=α−11Tr[eαlogρ+(1−α)logσ]
D_\alpha(\rho\|\sigma):= \frac{1}{\alpha-1}\log \text{Tr}\rho^\alpha \sigma^{(1-\alpha)} \leq D^\flat_\alpha(\rho\|\sigma):=\frac{1}{\alpha-1}\text{Tr}[e^{\alpha\log\rho+(1-\alpha)\log\sigma}]
Cheng, Hsieh, and Tomamichel. arXiv: 1704.05703
Dalai's Sphere-Packing Bound
logεn(R)1≤nEsp(R)+O(n)
\log \frac{1}{\varepsilon_n(R)} \leq n E_{\text{sp}}(R) + O(\sqrt{n})
Dalai. IEEE TIT, 59(12):8027–8056, 2013.
Shannon, Gallager, and Berlekamp. Information and Control, 10(1):65–103, 1967.
Theorem:
logεn(R)1≤nEsp(R)+O(logn)
\log \frac{1}{\varepsilon_n(R)} \leq n E_{\text{sp}}(R) + O(\log{n})
Cheng, Hsieh, and Tomamichel. arXiv: 1704.05703, 2017.
Altug and Wagner, IEEE TIT, 60(3): 1592–1614, 2014.
Proof:
Step 1：
εmax(Cn)≥σmaxxn∈Cnminα~∣Cn∣1(Wxn∣∣σ).
Step 2：
Two one-shot converse Hoeffding bounds for α~μ(⋅∣⋅).
α~μ(ρ∥σ)=minΠ{α(Π,ρ):β(Π,σ)≤μ}
\tilde{\alpha}_\mu(\rho\|\sigma) = \min_\Pi\{\alpha(\Pi,\rho): \beta(\Pi,\sigma)\leq \mu\}
For Bad Codewords, Use Weak Converse Hoeffding Bound.
α~e−nR(ρn∥σn)≥κ1e−κ2n−nϕn(R′∣ρn∥σn)
\tilde{\alpha}_{e^{-nR}}(\rho^n\|\sigma^n) \geq \kappa_1 e^{-\kappa_2\sqrt{n}-n\phi_n(R'|\rho^n\|\sigma^n)}
ϕn(r∣ρn∥σn):=supα∈(0,1]{α1−α(n1Dα(ρn∥σn)−r)}
\phi_n(r|\rho^n\|\sigma^n):= \sup_{\alpha\in(0,1]} \left\{\frac{1-\alpha}{\alpha}(\frac{1}{n}D_\alpha(\rho^n\|\sigma^n)-r)\right\}
H0:ρn=ρ1⊗⋯⊗ρn; H1:σn=σ1⊗⋯⊗σn
Blahut. IEEE TIT, 20(4):405–417, 1974.
Audenaert et. al., PRL 98:160501, 2007.
For Good Codewords, Use Sharp Converse Hoeffding Bound.
α~e−nR(ρn∥σn)≥n−tAe−nϕn(R′∣ρn∥σn)
\tilde{\alpha}_{e^{-nR}}(\rho^n\|\sigma^n) \geq \frac{A}{n^{-t}} e^{-n\phi_n(R'|\rho^n\|\sigma^n)}
H0:ρn=ρ1⊗⋯⊗ρn, H1:σn=σ1⊗⋯⊗σn
Bahadur and Rao, The Annals of Mathematical Statistics, 31(4):1015–1027, 1960.
Altug and Wagner, IEEE TIT, 60(3): 1592–1614, 2014.
t>1/2
t> 1/2
If Channel is Symmetric
Wx=Vx−1W1(V†)x−1,
W_x = V^{x-1}W_1 (V^\dagger)^{x-1},
logεn(R)1≤nEsp(R)+21(1+∣Esp′(R)∣)logn+o(1)
\log \frac{1}{\varepsilon_n(R)} \leq n E_{\text{sp}}(R) + \frac{1}{2}(1+|E'_{\text{sp}}(R)|)\log{n}+o(1)
then Sphere-packing bound is exact.
Property of E(R):
Iα(1)(P,W):=infσDα(P∘W∥P⊗σ)
I_{\alpha}^{(1)}(P,W):= \inf_\sigma D_\alpha(P\circ W\|P\otimes\sigma)
Iα(2)(P,W):=infσDα(W∥σ∣P)
I_{\alpha}^{(2)}(P,W):= \inf_\sigma D_\alpha(W\|\sigma|P)
(a) The map (α,P)→Iα is continuous on [0,1]×P(X).
(b) The map α→Iα is monotone increasing on [0,1].
(c) The map α→α1−αIα is strictly concave on (0,1].
Property of E(R):
Esp(1)(R,P):=sup0<α≤1α1−α(Iα(1)(P,W)−R)
E^{(1)}_{\text{sp}}(R,P):= \sup_{0<\alpha\leq 1} \frac{1-\alpha}{\alpha} \left(I_{\alpha}^{(1)}(P,W)-R\right)
(a) The map R→Esp(⋅) is convex, continuous, and non-increasing.
(b) Esp(⋅) is differentiable w.r.t. R.
(c) E′sp(⋅) is continuous.
Esp(2)(R,P):=sup0<α≤1α1−α(Iα(2)(P,W)−R)
E^{(2)}_{\text{sp}}(R,P):= \sup_{0<\alpha\leq 1} \frac{1-\alpha}{\alpha} \left(I_{\alpha}^{(2)}(P,W)-R\right)
Esp(R,P):=sups>0(E0(s,P)−sR)
E_{\text{sp}}(R,P):= \sup_{s>0} \left(E_{0}(s,P)-sR\right)
Part I.B
Moderate Deviation
Moderate Deviation:
εn(Rn)=e−Θ(nan2)
\varepsilon_n(R_n) = e^{-\Theta(na_n^2)}
{an}:an→0, ann→∞.
\{a_n\}: a_n\to 0,\ a_n\sqrt{n} \to \infty.
Rn=C−an,
R_n= C- a_n,
Cheng and Hsieh, arXiv: 1701.03195.
[Achievability] limsupn→∞nan21logεn(Rn)≤−2VW1
[Converse] liminfn→∞nan21logεn(Rn)≥−2VW1
Chubb, Tomamichel and Tan, arXiv: 1701.03114.
Achievability:
Step 1:
εn(Rn)≤4exp(−n[0≤s≤1maxE~0(s,P)−sRn])
Hayashi, PRA 76(6): 06230,12007.
nan21logεn(Rn)≤nan24−an21maxs{E~0(s,P)−sR}
\frac{1}{n a_n^2}\log\varepsilon_n(R_n)\leq \frac{4}{n a_n^2} - \frac{1}{a_n^2}\max_{s} \left\{ \tilde{E}_0(s,P)-sR\right\}
Achievability:
Step 2:
Apply Taylor Expansion to E~0(s,P) at s=0.
E~0(s,P)=sCW−2s2VW+6s3∂s3∂3E~0(s,P)∣s=sˉ
\tilde{E}_0(s,P) = s C_W - \frac{s^2}{2} V_W + \frac{s^3}{6} \frac{\partial^3 \tilde{E}_0(s,P)}{\partial s^3}|_{s=\bar{s}}
Property of E~0(s,P):
E~0(s,P):=−log∑xPxTrWx1−s(PW)s
\tilde{E}_0(s,P):= -\log \sum_x P_x\text{Tr} W_x^{1-s}(PW)^s
(a) Partial derivatives of E~0 are continuous.
(b) E~0 is concave in s≥0.
(c)∂s∂E~0(s,P)s=0=I(P,W).
\text{(c)}\left.\frac{\partial}{\partial s}\tilde{E}_0(s,P)\right|_{s=0} = I(P,W).
(d)∂s2∂2E~0(s,P)s=0=V(P,W).
\text{(d)}\left.\frac{\partial^2}{\partial s^2}\tilde{E}_0(s,P)\right|_{s=0} = V(P,W).
Converse
Similar to Quantum SP Bounds
1. A New Sharp Hoeffiding Bound.
2. Weak Hoeffiding Bound needs special attention.
New Sharp Converse Hoeffding Bound
α~e−nRn(ρn∥σn)≥sn⋆n−1/2Ae−nϕn(Rn′∣ρn∥σn)
\tilde{\alpha}_{e^{-nR_n}}(\rho^n\|\sigma^n) \geq \frac{A}{{s_n^\star} n^{-1/2}} e^{-n\phi_n(R_n'|\rho^n\|\sigma^n)}
H0:ρn=ρ1⊗⋯⊗ρn, H1:σn=σ1⊗⋯⊗σn
Chaganty-Sethuraman , The Annals of Probability, 21(3):1671–1690, 1993.
α~e−nR(ρn∥σn)≥n−tAe−nϕn(R′∣ρn∥σn),t>1/2
\tilde{\alpha}_{e^{-nR}}(\rho^n\|\sigma^n) \geq \frac{A}{ n^{-t}} e^{-n\phi_n(R'|\rho^n\|\sigma^n)}, \quad t>1/2
Summary
Part II
Source Coding (with quantum side information)
Xn
X^n
Bn
B^n
X^n
\hat{X}^n
En
\mathcal{E}_n
Dn
\mathcal{D}_n
Wn
W_n
εn(R)=Pr(Xn=X^n): ∣Wn∣=2nR
\varepsilon_n(R) = \text{Pr}(X^n\neq \hat{X}^n):\ |W_n| = 2^{nR}
limn→∞n−1logεn(R)=E(R)
\lim_{n\to\infty} \frac{-1}{n} \log\varepsilon_n(R) = E(R)
Error Exponent Analysis
Erc(R)≤E(R)≤Esp(R)
E_{\text{rc}}(R) \leq E(R) \leq E_{\text{sp}}(R)
Shannon, Bell System Technical Journal, 38(3):611–656, 1959.
Erc(R)=sup0.5≤α<1α1−α(R−H2−α1(X∣B)ρ)
E_{\text{rc}}(R)=\sup_{0.5\leq \alpha<1} \frac{1-\alpha}{\alpha}\left(R-H_{2-\frac{1}{\alpha}}(X|B)_\rho\right)
Theorem
Hα(X∣B)ρ=−Dα(ρXB∥IX⊗ρB)
H_\alpha(X|B)_\rho = -D_{\alpha}(\rho_{XB}\|\mathbb{I}_X\otimes \rho_B)
Esp(R)=sup0≤α≤1α1−α(R−Hα∗(X∣B)ρ)
E_{\text{sp}}(R)=\sup_{0\leq \alpha\leq1} \frac{1-\alpha}{\alpha}\left(R-H^*_{\alpha}(X|B)_\rho\right)
Hα∗(X∣B)ρ=maxσB−Dα(ρXB∥IX⊗σB)
H^*_\alpha(X|B)_\rho = \max_{\sigma_B}-D_{\alpha}(\rho_{XB}\|\mathbb{I}_X\otimes \sigma_B)
Theorem
R<CS
R< C_S
limn→∞n−1log(1−εn(R))=Esc(R)
\lim_{n\to\infty} \frac{-1}{n} \log\left(1-\varepsilon_n(R)\right) = E_{\text{sc}}(R)
Esc(R)=supα>1α1−α(R−Hαs(X∣B)ρ)
E_{\text{sc}}(R)=\sup_{ \alpha>1} \frac{1-\alpha}{\alpha}\left(R-H^s_{\alpha}(X|B)_\rho\right)
Hαs(X∣B)ρ=maxσB−Dαs(ρXB∥IX⊗σB)
H^s_\alpha(X|B)_\rho = \max_{\sigma_B}-D^s_{\alpha}(\rho_{XB}\|\mathbb{I}_X\otimes \sigma_B)
Theorem
Dαs(ρ∥σ)=α−11logTr[(ρ21σ2α1−αρ21)α]
D^s_{\alpha}(\rho\|\sigma)= \frac{1}{\alpha-1} \log\text{Tr}\left[\left(\rho^{\frac{1}{2}}\sigma^{\frac{1-\alpha}{2\alpha}}\rho^{\frac{1}{2}}\right)^\alpha\right]
Moderate Deviation:
limn→∞nan21logεn(Rn)=−2V(X∣B)ρ1
\lim_{n\to\infty} \frac{1}{na_n^2}\log\varepsilon_n(R_n) = - \frac{1}{2V(X|B)_\rho}
{an}:an→0, ann→∞.
\{a_n\}: a_n\to 0,\ a_n\sqrt{n} \to \infty.
Rn=CS+an
R_n= C_S+ a_n
Rn(εn)=CS+2V(X∣B)ρan+O(an)
R_n(\varepsilon_n) = C_S + \sqrt{2V(X|B)_\rho} a_n + O(a_n)
εn=e−nan2
\varepsilon_n= e^{-na_n^2}
Duality
Esp,source(R,QX)=Esp,channel(H(QX)−R,QX)
E_{\text{sp,source}}(R,Q_X) = E_{\text{sp,channel}}(H(Q_X)-R,Q_X)
Open Questions
1. Beyond C-Q Channel?
2. EA Channel?
3. Duality?
Thank you!
Error Exponent Analysis in Quantum Source and Channel Coding 1. [ arXiv:1803.07505 ]: Cheng (NTU&UTS), Hanson (Cambridge), Datta (Cambridge), MH 3. [ arXiv:1701.03195 ]: Cheng (NTU&UTS), MH 2. [ arXiv:1704.05703 ]: Cheng (NTU&UTS), MH , Tomamichel (UTS)
Error Exponent Analysis in Quantum Source and Channel Coding
By Lawrence Min-Hsiu Hsieh
Error Exponent Analysis in Quantum Source and Channel Coding
- 596
