On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf ·...

61
On some subadditivity inequalities of entropies AdamBESENYEI Dept. of Applied Analysis Eotv os Lor and University, Budapest, Hungary [email protected] April 12, 2014 A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 1 / 18

Transcript of On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf ·...

Page 1: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

On some subadditivity inequalities of entropies

Adam BESENYEI

Dept. of Applied AnalysisEotvos Lorand University, Budapest, Hungary

[email protected]

April 12, 2014

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 1 / 18

Page 2: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Outline

1 Entropy in probability• Shannon entropy• Tsallis entropy

}origins and their subadditivity

2 Quantum entropy• von Neumann entropy• quantum Tsallis entropy

}origins and their subadditivity

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 2 / 18

Page 3: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

Problem

Let X be a discrete random variable with possible values of outcome {x1, . . . , xm}and probability distribution p(xi ) = P(X = xi ).How to measure the “uncertainty” of the random variable?

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 4: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

Problem

Let X be a discrete random variable with possible values of outcome {x1, . . . , xm}and probability distribution p(xi ) = P(X = xi ).How to measure the “uncertainty” of the random variable?

Example

Coin toss (not fair): P(head) = p, P(tail) = 1− p.If p = 0 or p = 1, there is no uncertainty.As p increases from 0 to 1

2 , the uncertainty also increases.If p = 1

2 , then the uncertainty is maximal.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 5: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

Definition

Let X be a discrete random variable with possible values of outcome {x1, . . . , xm}and probability distribution p(x) = P(X = x). Then, the Shannon entropy of thediscrete variable X is defined as

H(X ) = −m∑i=1

p(xi ) log p(xi )

with the convention that 0 log 0 = 0, since limx→0+

x log x = 0.

(The letter H is the Greek capital eta.)

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 6: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

History

Entropy (“turning towards”) in classical thermodynamics was introduced in 1865by Rudolph Clausius (1822–1888), later by Josiah Willard Gibbs (1839–1903), andby Ludwig Boltzmann (1844–1906) in a statistical point of view.

Rudolph Clausius Willard Gibbs Ludwig Boltzmann

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 7: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

History

The formula in information theory was introduced by Claude Shannon(1916–2001) in his seminal paper A Mathematical Theory of Communication in1948 while he was working at Bell Telephone.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 8: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 9: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

History

The paper was reprinted in 1949 in the book The Mathematical Theory ofCommunication by Claude Shannon and Warren Weaver (1894–1978).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 10: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Shannon entropy

Shannon entropy: definition and origins

How to call it?Shannon visited John von Neumann in 1949. According to some sources,Neumann said:

“You should call it entropy, for two reasons. In the first place youruncertainty function has been used in statistical mechanics under thatname, so it already has a name. In the second place, and moreimportant, nobody knows what entropy really is, so in a debate you willalways have the advantage.”

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 3 / 18

Page 11: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Shannon entropy

DefinitionIf Y is another discrete random variable with possible values of outcome{y1, . . . , yn} and probability distribution p(y), then the joint entropy of the pair(X ,Y ) with joint distribution p(x , y) is

H(X ,Y ) = −m∑i=1

n∑j=1

p(xi , yj) log p(xi , yj).

TheoremThe joint entropy is not greater than the sum of the individual entropies:

H(X ,Y ) ≤ H(X ) + H(Y ).

This is called the subadditivity of the entropy.

RemarkEquality holds if and only if X and Y are independent.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 4 / 18

Page 12: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Shannon entropy

DefinitionIf Y is another discrete random variable with possible values of outcome{y1, . . . , yn} and probability distribution p(y), then the joint entropy of the pair(X ,Y ) with joint distribution p(x , y) is

H(X ,Y ) = −m∑i=1

n∑j=1

p(xi , yj) log p(xi , yj).

TheoremThe joint entropy is not greater than the sum of the individual entropies:

H(X ,Y ) ≤ H(X ) + H(Y ).

This is called the subadditivity of the entropy.

RemarkEquality holds if and only if X and Y are independent.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 4 / 18

Page 13: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Shannon entropy

DefinitionIf Y is another discrete random variable with possible values of outcome{y1, . . . , yn} and probability distribution p(y), then the joint entropy of the pair(X ,Y ) with joint distribution p(x , y) is

H(X ,Y ) = −m∑i=1

n∑j=1

p(xi , yj) log p(xi , yj).

TheoremThe joint entropy is not greater than the sum of the individual entropies:

H(X ,Y ) ≤ H(X ) + H(Y ).

This is called the subadditivity of the entropy.

RemarkEquality holds if and only if X and Y are independent.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 4 / 18

Page 14: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Shannon entropy

TheoremIf X ,Y ,Z are discrete random variables, then

H(X ,Y ,Z ) + H(Y ) ≤ H(X ,Y ) + H(Y ,Z ).

This is called the strong subadditivity of the entropy.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 5 / 18

Page 15: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Shannon entropy

TheoremIf X ,Y ,Z are discrete random variables, then

H(X ,Y ,Z ) + H(Y ) ≤ H(X ,Y ) + H(Y ,Z ).

This is called the strong subadditivity of the entropy.

RemarkBy introducing the conditional entropies

H(Z |X ,Y ) = H(X ,Y ,Z )− H(X ,Y )

H(Z |Y ) = H(Y ,Z )− H(Y ),

the strong subadditivity inequality can be written as

H(Z |X ,Y ) ≤ H(Z |Y ).

Equality holds if and only if X ,Y ,Z form a Markov-triplet, i.e.,P(Z |X ,Y ) = P(Z |Y ) (X = “past”,Y = “present”,Z = “future”).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 5 / 18

Page 16: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Shannon entropy

TheoremIf X ,Y ,Z are discrete random variables, then

H(X ,Y ,Z ) + H(Y ) ≤ H(X ,Y ) + H(Y ,Z ).

This is called the strong subadditivity of the entropy.

Some notationLet us denote the joint distribution of X ,Y ,Z by

{pijk := p(xi , yj , zk) : 1 ≤ i ≤ m, 1 ≤ j ≤ n, 1 ≤ k ≤ r},and the marginal distributions p(xi , yj), p(yj , zk), p(yj) by

pij− :=r∑

k=1

pijk , p−jk :=m∑i=1

pijk , p−j− :=m∑i=1

r∑k=1

pijk .

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 5 / 18

Page 17: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Shannon entropy

TheoremIf X ,Y ,Z are discrete random variables, then

H(X ,Y ,Z ) + H(Y ) ≤ H(X ,Y ) + H(Y ,Z ).

This is called the strong subadditivity of the entropy.

A reformulationThen the strong subadditivity has the form∑

ijk

−pijk log pijk −∑j

p−j− log p−j− ≤ −∑ij

pij− log pij− −∑jk

p−jk log p−jk ,

or equivalently∑ijk

pijk (log pijk + log p−j− − log pij− − log p−jk) ≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 5 / 18

Page 18: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Shannon entropy

TheoremFor each fixed 1 ≤ j ≤ n and 1 ≤ k ≤ r the following inequality holds:

n∑i=1

pijk (log pijk + log p−j− − log pij− − log p−jk) ≥ 0.

Proof.The inequality is equivalent to∑

i

pijkp−jk

log

(pij−p−jkpijkp−j−

)≤ 0 where

∑i

pijkp−jk

= 1.

The concavity of the logarithm function implies∑i

pijkp−jk

log

(pij−p−jkpijkp−j−

)≤ log

(∑i

pijkp−jk

pij−p−jkpijkp−j−

)= log 1 = 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 6 / 18

Page 19: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Shannon entropy

TheoremFor each fixed 1 ≤ j ≤ n and 1 ≤ k ≤ r the following inequality holds:

n∑i=1

pijk (log pijk + log p−j− − log pij− − log p−jk) ≥ 0.

Proof.The inequality is equivalent to∑

i

pijkp−jk

log

(pij−p−jkpijkp−j−

)≤ 0 where

∑i

pijkp−jk

= 1.

The concavity of the logarithm function implies∑i

pijkp−jk

log

(pij−p−jkpijkp−j−

)≤ log

(∑i

pijkp−jk

pij−p−jkpijkp−j−

)= log 1 = 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 6 / 18

Page 20: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Tsallis entropy

Tsallis entropy: definition and origins

DefinitionIf X is a discrete random variable and q 6= 1, then the Tsallis entropy or q-entropyis defined as

Sq(X ) =1−

∑mi=1 p(xi )

q

q − 1.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 7 / 18

Page 21: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Tsallis entropy

Tsallis entropy: definition and origins

DefinitionIf X is a discrete random variable and q 6= 1, then the Tsallis entropy or q-entropyis defined as

Sq(X ) =1−

∑mi=1 p(xi )

q

q − 1.

RemarkBy introducing the q-logarithm function

logq x =xq−1 − 1

q − 1(q 6= 1),

the Tsallis entropy can be written as

Sq(X ) = −m∑i=1

p(xi ) logq p(xi ).

It is easily seen that limq→1

Sq = H.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 7 / 18

Page 22: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Tsallis entropy

Tsallis entropy: definition and origins

History

This generalized entropy was introduced by Constantino Tsallis (1943–) in hispaper A possible generalization of Boltzmann–Gibbs statistics in 1988.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 7 / 18

Page 23: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Tsallis entropy

Tsallis entropy: definition and origins

History

However, a similar entropy was already defined by Daroczy Zoltan (1938–) in hispaper Generalized information functions in 1970.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 7 / 18

Page 24: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Tsallis entropy

Theorem (Daroczy, 1970)

If q > 1, then the q-entropy is subadditive:

Sq(X ,Y ) ≤ Sq(X ) + Sq(Y ).

Equality holds if and only if q = 1 and X ,Y are independent.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 8 / 18

Page 25: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Tsallis entropy

Theorem (Daroczy, 1970)

If q > 1, then the q-entropy is subadditive:

Sq(X ,Y ) ≤ Sq(X ) + Sq(Y ).

Equality holds if and only if q = 1 and X ,Y are independent.

A special case (proposed in KoMaL, 2013)

For m = n = 2, the subadditivity reduces to the nice inequality

(x + y)q + (z + v)q + (x + z)q + (y + v)q ≤ xq + yq + zq + vq + (x + y + z + v)q.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 8 / 18

Page 26: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Tsallis entropy

Theorem (Daroczy, 1970)

If q > 1, then the q-entropy is subadditive:

Sq(X ,Y ) ≤ Sq(X ) + Sq(Y ).

Equality holds if and only if q = 1 and X ,Y are independent.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 8 / 18

Page 27: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Tsallis entropy

Theorem (Daroczy, 1970)

If q > 1, then the q-entropy is subadditive:

Sq(X ,Y ) ≤ Sq(X ) + Sq(Y ).

Equality holds if and only if q = 1 and X ,Y are independent.

A similar inequality (proposed in Amer. Math. Monthly, 2014)

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 8 / 18

Page 28: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Subadditivity of Tsallis entropy

Proof (Case m = n = 2).

Let f (x) = x − xq, then f is non-negative and convex. The subadditivity is

f (x) + f (y) + f (z) + f (v) ≤ f (x + y) + f (z + v) + f (x + z) + f (y + v).

Observe

f (x) + f (y) + f (z) + f (v)− f (x + y)− f (z + v)

= (x + y)q[f

(x

x + y

)+ f

(y

x + y

)]+ (z + v)q

[f

(z

z + v

)+ f

(v

z + v

)].

Since (x + y) + (z + v) = 1 and (x + y)q ≤ x + y , (z + v)q ≤ z + v , thereforeJensen’s inequality implies

(x + y)qf

(x

x + y

)+ (z + v)qf

(z

z + v

)≤ f (x + z),

(x + y)qf

(y

x + y

)+ (z + v)qf

(v

z + v

)≤ f (y + v)

thus the subadditivity inequality follows.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 8 / 18

Page 29: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Tsallis entropy

Theorem (Shigeru Furuichi, 2004)

If q > 1, then the q-entropy is strongly subadditive:

Sq(X ,Y ,Z ) + Sq(Y ) ≤ Sq(X ,Y ) + Sq(Y ,Z ).

Equality holds if and only if q = 1 and X ,Y ,Z form a Markov triplet.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 9 / 18

Page 30: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Strong subadditivity of Tsallis entropy

Theorem (Shigeru Furuichi, 2004)

If q > 1, then the q-entropy is strongly subadditive:

Sq(X ,Y ,Z ) + Sq(Y ) ≤ Sq(X ,Y ) + Sq(Y ,Z ).

Equality holds if and only if q = 1 and X ,Y ,Z form a Markov triplet.

RemarkAs q → 1+, we obtain the subadditivity inequalities of the Shannon-entropy.The strong subadditivity can be reformulated as∑

ijk

pijk(logq pijk + logq p−j− − logq pij− − logq p−jk

)≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 9 / 18

Page 31: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Tsallis entropy

Theorem (B.–Petz, 2013)

For q > 1 and fixed 1 ≤ j ≤ n, 1 ≤ k ≤ r the following inequality holds:n∑

i=1

pijk(logq pijk + logq p−j− − logq pij− − logq p−jk

)≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 10 / 18

Page 32: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Tsallis entropy

Theorem (B.–Petz, 2013)

For q > 1 and fixed 1 ≤ j ≤ n, 1 ≤ k ≤ r the following inequality holds:n∑

i=1

pijk(logq pijk + logq p−j− − logq pij− − logq p−jk

)≥ 0.

Lemma

Let ai , bi (i = 1, . . . ,m) be positive numbers and q ≥ 1. Then

m∑i=1

aibi (ai + bi )q−2 ≤

m∑i=1

ai ·m∑i=1

bi ·

(m∑i=1

(ai + bi )

)q−2

.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 10 / 18

Page 33: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Tsallis entropy

Theorem (B.–Petz, 2013)

For q > 1 and fixed 1 ≤ j ≤ n, 1 ≤ k ≤ r the following inequality holds:n∑

i=1

pijk(logq pijk + logq p−j− − logq pij− − logq p−jk

)≥ 0.

Lemma

Let ai , bi (i = 1, . . . ,m) be positive numbers and q ≥ 1. Then

m∑i=1

aibi (ai + bi )q−2 ≤

m∑i=1

ai ·m∑i=1

bi ·

(m∑i=1

(ai + bi )

)q−2

.

Remark

For q = 1, this is called Milne’s inequality (1925). (Edward Arthur Milne(1896–1950) British astrophysicist and mathematician.)

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 10 / 18

Page 34: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Entropy in probability Subadditivity inequalities

Partial strong subadditivity of Tsallis entropy

Proof.The inequality is equivalent to

n∑i=1

pijk(pq−1ij− − pq−1

ijk

)≤ p−jk

(pq−1−j− − pq−1

−jk

).

Denote ai = pijk , bi = pij− − pijk (i = 1, . . . ,m) and A =∑m

i=1 ai , B =∑m

i=1 bi .Then A = p−jk and A + B = p−j− so the inequality reduces to the form

m∑i=1

ai(

(ai + bi )q−1 − aq−1

i

)≤ A

((A + B)q−1 − Aq−1

),

or equivalently

(q − 1)

∫ 1

0

m∑i=1

aibi (ai + tbi )q−2 dt ≤ (q − 1)

∫ 1

0

AB(A + tB)q−2 dt,

which follows from the Lemma.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 10 / 18

Page 35: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy von Neumann entropy

Generalization to density matrices

von Neumann entropy

Instead of a discrete probability distribution, we have a density matrix:

ρ ≥ 0, Tr ρ = 1.

Then the von Neumann entropy is defined as

S(ρ) = −Tr(ρ log ρ).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 11 / 18

Page 36: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy von Neumann entropy

Generalization to density matrices

von Neumann entropy

Instead of a discrete probability distribution, we have a density matrix:

ρ ≥ 0, Tr ρ = 1.

Then the von Neumann entropy is defined as

S(ρ) = −Tr(ρ log ρ).

Remark

If ρ has eigenvalues λi (i = 1, . . . , n), then

Sq(ρ) = −n∑

i=1

λi log λi .

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 11 / 18

Page 37: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy von Neumann entropy

Generalization to density matrices

History

This formula was first studied by John von Neumann (1903–1957) in his paperThermodynamik quantenmechanischer Gesamtheiten in 1927 and later in hisfamous book Mathematische Grundlagen der Quantenmechanik in 1932.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 11 / 18

Page 38: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy von Neumann entropy

Generalization to density matrices

History

This formula was first studied by John von Neumann (1903–1957) in his paperThermodynamik quantenmechanischer Gesamtheiten in 1927 and later in hisfamous book Mathematische Grundlagen der Quantenmechanik in 1932.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 11 / 18

Page 39: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

Partial traces

Let X ∈Mm ⊗Mn, X =∑

i Ai ⊗ Bi . Then the partial traces Tr1 X ∈Mn andTr2 X ∈Mm are defined as

Tr1 X =∑i

Tr(Ai ) · Bi , Tr2 X =∑i

Tr(Bi ) · Ai .

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 40: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

Partial traces

Let X ∈Mm ⊗Mn, X =∑

i Ai ⊗ Bi . Then the partial traces Tr1 X ∈Mn andTr2 X ∈Mm are defined as

Tr1 X =∑i

Tr(Ai ) · Bi , Tr2 X =∑i

Tr(Bi ) · Ai .

RemarkThe partial trace Tr1 is the adjoint of the canonical embedding

Mn 3 B 7→ Im ⊗ B ∈Mm ⊗Mn.

Similarly, Tr2 is the adjoint of the canonical embedding

Mm 3 A 7→ A⊗ In ∈Mm ⊗Mn.

If Eij ∈Mm (i , j = 1, 2, . . . ,m) are the matrix units, then X =∑m

i,j=1 Eij ⊗Xij , so

Tr1 X =m∑i=1

Xii ∈Mn, Tr2 X = [TrXij ]mi,j=1 ∈Mm.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 41: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

Partial traces

Let X ∈Mm ⊗Mn, X =∑

i Ai ⊗ Bi . Then the partial traces Tr1 X ∈Mn andTr2 X ∈Mm are defined as

Tr1 X =∑i

Tr(Ai ) · Bi , Tr2 X =∑i

Tr(Bi ) · Ai .

RemarkIf ρ12 ∈Mm ⊗Mn is a density matrix, then its reduced densities are

ρ1 = Tr2 ρ12 ∈Mm, ρ2 = Tr1 ρ12 ∈Mn.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 42: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

TheoremIf ρ12 ∈Mm ⊗Mn is a density matrix with reduced densities ρ1 = Tr2 ρ12 ∈Mm,ρ2 = Tr1 ρ12 ∈Mn, then the subadditivity inequality holds true:

S(ρ12) ≤ S(ρ1) + S(ρ2).

Equality holds if and only if ρ12 = ρ1 ⊗ ρ2.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 43: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

TheoremIf ρ12 ∈Mm ⊗Mn is a density matrix with reduced densities ρ1 = Tr2 ρ12 ∈Mm,ρ2 = Tr1 ρ12 ∈Mn, then the subadditivity inequality holds true:

S(ρ12) ≤ S(ρ1) + S(ρ2).

Equality holds if and only if ρ12 = ρ1 ⊗ ρ2.

Theorem (Lieb–Ruskai, 1973)

If ρ123 ∈Mm ⊗Mn ⊗Mr is a density matrix with reduced densitiesρ12 = Tr3 ρ123 ∈Mr , ρ23 = Tr1 ρ123 ∈Mm, ρ2 = Tr13 ρ123 ∈Mm ⊗Mr , then thestrong subadditivity inequality holds true:

S(ρ123) + S(ρ2) ≤ S(ρ12) + S(ρ23).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 44: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Strong subadditivity of von Neumann entropy

Elliott Hershel Lieb (1932–) Mary Beth Ruskai (1944–)

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 12 / 18

Page 45: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of von Neumann entropy

A reformulationThe strong subadditivity inequality can be written in the form

Tr(ρ123(log ρ123 + I1 ⊗ log ρ2 ⊗ I3 − log ρ12 ⊗ I3 − I1 ⊗ log ρ23)

)≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 13 / 18

Page 46: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of von Neumann entropy

A reformulationThe strong subadditivity inequality can be written in the form

Tr(ρ123(log ρ123 + I1 ⊗ log ρ2 ⊗ I3 − log ρ12 ⊗ I3 − I1 ⊗ log ρ23)

)≥ 0.

Theorem (Isaac Kim, 2012, B.–Petz, 2013)

For a density matrix ρ123 ∈Mm ⊗Mn ⊗Mr the following partial subadditivityinequality holds true:

Tr12

(ρ123(log ρ123 + I1 ⊗ log ρ2 ⊗ I3 − log ρ12 ⊗ I3 − I1 ⊗ log ρ23)

)≥ 0,

i.e., the matrix on the left-hand side is positive semidefinite.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 13 / 18

Page 47: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of von Neumann entropy

A reformulationThe strong subadditivity inequality can be written in the form

Tr(ρ123(log ρ123 + I1 ⊗ log ρ2 ⊗ I3 − log ρ12 ⊗ I3 − I1 ⊗ log ρ23)

)≥ 0.

Theorem (Isaac Kim, 2012, B.–Petz, 2013)

For a density matrix ρ123 ∈Mm ⊗Mn ⊗Mr the following partial subadditivityinequality holds true:

Tr12

(ρ123(log ρ123 + I1 ⊗ log ρ2 ⊗ I3 − log ρ12 ⊗ I3 − I1 ⊗ log ρ23)

)≥ 0,

i.e., the matrix on the left-hand side is positive semidefinite.

Idea of the proof

Use the joint convexity of the quasi-entropy

SK− log(P‖Q) = Tr(K∗KQ logQ − KQK∗ logP).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 13 / 18

Page 48: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Tsallis entropy

Generalization to density matrices

Tsallis entropy

For a density matrix ρ the quantum Tsallis entropy is defined as

Sq(ρ) =1− Tr ρq

q − 1= −Tr(ρ logq ρ)

where the q-logarithm function for q 6= 1 is given by

logq x =xq−1 − 1

q − 1(q 6= 1).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 14 / 18

Page 49: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Tsallis entropy

Generalization to density matrices

Tsallis entropy

For a density matrix ρ the quantum Tsallis entropy is defined as

Sq(ρ) =1− Tr ρq

q − 1= −Tr(ρ logq ρ)

where the q-logarithm function for q 6= 1 is given by

logq x =xq−1 − 1

q − 1(q 6= 1).

Remark

If ρ has eigenvalues λi (i = 1, . . . , n), then

Sq(ρ) =1−

∑ni=1 λ

qi

q − 1.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 14 / 18

Page 50: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

Theorem (Koenraad Audenaert, 2007)

If ρ12 ∈Mm ⊗Mn is a density matrix with reduced densities ρ1 = Tr2 ρ12 ∈Mm

and ρ2 = Tr1 ρ12 ∈Mn, then for q > 1 the subadditivity inequality holds true:

Sq(ρ12) ≤ Sq(ρ1) + Sq(ρ2),

or equivalently

Tr ρq1 + Tr ρq2 ≤ Tr ρq12.

Equality holds if and only if ρ12 = ρ1 ⊗ ρ2 where min(rank ρ1, rank ρ2) = 1.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 51: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

RemarkThe strong subadditivity is not true in general. For example, if

ρ123 =

0.02 0 0 0 0 0 0 00 0.02 0 0 0 0 0 00 0 0.4 0 0.12 0 0 00 0 0 0.06 0 0.12 0 00 0 0.12 0 0.06 0 0 00 0 0 0.12 0 0.4 0 00 0 0 0 0 0 0.02 00 0 0 0 0 0 0 0.02

,

then

S2(ρ12) + S2(ρ23)− S2(ρ123)− S2(ρ2) = −0.0108.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 52: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

A nice special case

If q = 2 and

ρ12 =

[A BB∗ C

]∈M2 ⊗Mn,

then

ρ1 =

[TrA TrB

TrB∗ TrC

], ρ2 = A + C ,

therefore the subadditivity inequality reduces to the neat inequality

TrAC − TrB∗B ≤ TrATrC − TrB∗ TrB.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 53: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

A nice special case

If q = 2 and

ρ12 =

[A BB∗ C

]∈M2 ⊗Mn,

then

ρ1 =

[TrA TrB

TrB∗ TrC

], ρ2 = A + C ,

therefore the subadditivity inequality reduces to the neat inequality

TrAC − TrB∗B ≤ TrATrC − TrB∗ TrB.

Remark

Equality holds if and only if ρ12 = ρ1 ⊗ ρ2 such that min(rank ρ1, rank ρ2) ≤ 1.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 54: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

A nice special case

If q = 2 and

ρ12 =

[A BB∗ C

]∈M2 ⊗Mn,

then

ρ1 =

[TrA TrB

TrB∗ TrC

], ρ2 = A + C ,

therefore the subadditivity inequality reduces to the neat inequality

TrAC − TrB∗B ≤ TrATrC − TrB∗ TrB.

RemarkElementary proof given in the Bulletin of the International Linear Algebra Society.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 55: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

Proof.Due to unitary invariance, we may assume that A is diagonal. Then the inequalityreduces to the form

2∑i>j

<(biibjj)−∑i 6=j

|bij |2 ≤∑i>j

(aiicjj + ajjcii ).

Since

[A BB∗ C

]≥ 0, therefore A ≥ 0, C ≥ 0 and the 2x2 principal

subdeterminants of X are also non-negative:· · · · ·· aii · bii ·· · · · ·· bii · cii ·· · · · ·

thus aiicii ≥ |bii |2 for i = 1, . . . , n. Therefore, by the AGM inequality,

aiicjj + ajjcii ≥ 2√aiiciiajjcjj ≥ 2

√|bii |2|bjj |2 ≥ 2<(biibjj).

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 56: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Subadditivity of Tsallis entropy

Remark

A generalization was shown by Tsuyoshi Ando (1932–):[TrC · A + BB∗ −AC − TrB∗ · B−CA− TrB · B∗ TrA · C + B∗B

]≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 15 / 18

Page 57: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of Tsallis entropy

A reformulationThe subadditivity inequality for the quantum Tsallis entropy can be written as

Tr(ρ12(logq ρ12 − logq ρ1 ⊗ I2 − I1 ⊗ logq ρ2)

)≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 16 / 18

Page 58: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of Tsallis entropy

A reformulationThe subadditivity inequality for the quantum Tsallis entropy can be written as

Tr(ρ12(logq ρ12 − logq ρ1 ⊗ I2 − I1 ⊗ logq ρ2)

)≥ 0.

Open problem

For a density matrix ρ12 ∈Mm ⊗Mn the following partial subadditivity inequalitymight hold true:

Tr1

(ρ12(logq ρ12 − logq ρ1 ⊗ I2 − I1 ⊗ logq ρ2)

)≥ 0.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 16 / 18

Page 59: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

Quantum entropy Subadditivity inequalities

Partial subadditivity of Tsallis entropy

A reformulationThe subadditivity inequality for the quantum Tsallis entropy can be written as

Tr(ρ12(logq ρ12 − logq ρ1 ⊗ I2 − I1 ⊗ logq ρ2)

)≥ 0.

Open problem

For a density matrix ρ12 ∈Mm ⊗Mn the following partial subadditivity inequalitymight hold true:

Tr1

(ρ12(logq ρ12 − logq ρ1 ⊗ I2 − I1 ⊗ logq ρ2)

)≥ 0.

RemarkThe conjecture is true for

• ρ12 = ρ1 ⊗ ρ2 and q > 1,

• ρ12 ∈M2n ⊗M2n and q = 2.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 16 / 18

Page 60: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

References

T. Ando, Positivity of certain block-matrices, manuscript, 2013.

K. M. R. Audenaert, Subadditivity of q-entropies for q > 1, J. Math. Phys.,48 (2007), 083507.

Adam Besenyei, Denes Petz, Partial subadditivity of entropies, Linear AlgebraAppl., 439 (2013), 3297–3305.

Z. Daroczi, General information functions, Information and Control, 16(1970), 36–51.

S. Furuichi, Information theoretical properties of Tsallis entropies,J.Math.Phys., 47 (2006), 023302.

I. H. Kim, Operator extension of strong subadditivity of entropy, J. Math.Phys., 53 (2012), 122204.

E. H. Lieb, M. B. Ruskai, Proof of the strong subadditivity of quantummechanical entropy, J. Math. Phys., 14 (1973), 1938–1941.

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 17 / 18

Page 61: On some subadditivity inequalities of entropiesabesenyei.web.elte.hu/publications/entropy.pdf · CommunicationbyClaude ShannonandWarren Weaver(1894{1978). A. Besenyei (ELTE Budapest)

The End

Thank you for your attention!

A. Besenyei (ELTE Budapest) Subadditivity inequalities April 12, 2014 18 / 18