Computation of Total Capacity for Discrete Memoryless Multiple-Access Channels

Size: px
Start display at page:

Download "Computation of Total Capacity for Discrete Memoryless Multiple-Access Channels"

Transcription

1 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER Computation of Total Capacit for Discrete emorless ultiple-access Channels ohammad Rezaeian, ember, IEEE, and Ale Grant, Senior ember, IEEE Abstract The Arimoto Blahut algorithm is generalized for computation of the total capacit of discrete memorless multiple-access channels (ACs). In addition,a class of ACs is defined with the propert that the uniform distribution achieves the total capacit. These results are based on the specialization of the Kuhn Tucker condition for the total capacit of the AC,and an etension of a known smmetr propert for single-user channels. Inde Terms Arimoto Blahut algorithm,capacit,multiple-access channel (AC),nonconve optimization. I. INTRODUCTION Determination of the capacit region for multiterminal channels has attracted much attention in information theor. In man cases, single-letter representations of the capacit region are not known. Even in cases where a single-letter description has been found, such as the discrete memorless multiple-access channel (AC), evaluation of the capacit region is problematic. Specificall, computation of the boundar of the capacit region is a nonconve optimization problem. In contrast, for single-user channels in case of smmetr the capacit-achieving distribution is known, and in other cases, channel capacit can be numericall approimated to arbitrar precision using the Arimoto Blahut algorithm [1] [3] or other numerical optimization procedures [4], [5]. Such techniques are still lacking for the AC in its general form. A numerical method has been developed for two-user ACs with binar output [6]. Of particular interest for this correspondence is the computation of the total capacit, C total for the discrete memorless AC. This is the solution of the following optimization problem: ma I(X 1;X 2;...;X ; Y ): (1) P (X )P (X )...P (X ) The problem of capacit computation for a single-user channel is conve, and therefore the Kuhn Tucker condition [7] is sufficient for a distribution to achieve capacit. For a AC, this conveit is missing. Nevertheless, recent results [8] have shown that the Kuhn Tucker condition is either sufficient for optimalit in a AC, or the channel can be decomposed into subchannels for which the Kuhn Tucker condition is sufficient for optimalit. In the latter case, at least one subchannel has an optimal distribution that achieves the capacit of the original channel. In light of the result reported in [8], this correspondence gives capacit computation methods for the AC analogous to the methods used for single-user channels. Starting with an information function defined in Section II, a specialization of the Kuhn Tucker condition for the maimization (1) is given in Section III. This allows the derivation anuscript received August 7, 2003; revised a 27, This work was supported b the Australian Government under ARC Grant DP Rezaeian was with the Institute for Telecommunications Research, Universit of South Australia, awson Lakes SA 5095 Australia. He is now with the Department of Electrical and Electronic Engineering, Universit of elbourne, elbourne, Australia. A. Grant is with the Institute for Telecommunications Research, Universit of South Australia, awson Lakes SA 5095 Australia. Communicated b R. Y. Yeung, Associate Editor for Shannon Theor. Digital Object Identifier /TIT of a generalized Arimoto Blahut algorithm for total capacit, eplained in Section IV. Section V defines a class of smmetric -user discrete memorless ACs for which the uniform distribution is the solution of (1). This is based upon a generalization of the well-known notion of a smmetric single-user channel [10, p. 94]. Furthermore, a subset of this class in fact ehibits a special smmetr propert in the channel transition probabilit table. Notation P (Y ) and P (Y j X) represent probabilit distribution functions P Y (1) and P Y jx (1j1) (the latter being a conditional probabilit). Lower case arguments denote that the variable has been set to a specific value in the corresponding function, e.g., P (Y j ) represents P Y jx (1j). II. THE INFORATION FUNCTION A fundamental function, evaluating the information associated with individual elements of the discrete support X of a random variable X in the contet of the joint distribution P (X; Y ) is I(; Y )=D (P (Y j )kp (Y )) (2) where D(1k1) is the Kullback Leibler distance D P (Y )kp 0 (Y ) = P ()log P () P 0 () : Based on this definition, mutual information and entrop are given b the following epectations, taken with respect to P (X): where I(X; Y )= X[I(; Y )] (3) H(X) = X[I(; X)] (4) I(; X) =D(P (X j )kp (X)) = 0 log P () is the self-information associated with. The function I(; Y ) measures the D-variation in the distribution on Y due to revelation of. Whereas a distribution P (Y ) represents probabilistic knowledge of Y, its variation represents the information associated with a particular event. Gallager [10, p. 16] defines I(; ) log(p (j)=p ()) as the mutual information and (2) as an average mutual information over Y (with respect to P (Y j), rather than P (Y )). Now although I(; ) is not alwas positive, I(; Y ) 0 and in light of the preceding discussion (2) will be referred to as the information function. The information function can be etended to a set of variables X i 2X i, namel I( 1 ; 2 ; ::; ; Y )=D(P (Y j 1 ; 2 ;...; )kp (Y )): For S f1; 2;...;g, let S = f i : i 2Sgand denote the marginalization of I( 1 ; 2 ; ::; ; Y ) to S b IS (S; Y )= P S ( S )I(; Y ) where S = ns. In the case of a singleton set, notation will be abused to write I m = Ifmg. These notations should not be confused with I(S ; Y )=D(P (Y j S )kp (Y )). Note that I(X; Y )= X [IS (S; Y )] (5) I(XS; Y )= X [I(S ; Y )]: (6) /04$ IEEE

2 2780 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER 2004 The information function I(S ; Y ) depends on the joint distribution P (XS ; Y ). For fied conditional probabilit P (Y j XS ), the information function, and therefore mutual information, is a functional of the joint distribution P (XS). Each probabilit value P (S ) can therefore be considered as a variable and the mutual information I(XS; Y ) as a function of these variables. an information-theoretic problems are concerned with optimization of I(XS; Y ) over certain selections of variables P (S ). The first- and second-order derivatives of I(XS; Y ) Y ) = I(S ; Y ) I(XS; Y 0 S ) = 0 P ( j S )P ( j 0 S ) : (8) P () If however the random variables X are mutuall independent, with P (X) = m=1 P m(xm) then the mutual information I(X; Y ) can be regarded as a function of variables P m ( m ) for m 2X m, m 2, and it can be shown Y m(m) = I m ( m ; Y ) I(X; Y m(m)@pm( 0 m ) = 0 P ( j m)p ( j 0 m) P 2 I(X; Y m ( m )@P m ( m ) = I fm;m g( m;m ; Y ) 0 P ( j m )P ( j m ) (10) : (11) P () The derivatives (7) and (8) are along the lines of [10, eq. (4.5.5)], whereas the relations (9) (11) appear to be new. The first-order derivative is used to obtain the Kuhn Tucker condition for a distribution that maimizes mutual information. The second derivative can be used to obtain the Hessian matri for the mutual information at an given distribution, which could be used for conveit analsis of the mutual information although this approach is not pursued here. For a pair of random variables X; Y with fied P (Y j X), mutual information I(X; Y ) is a concave function of the variables P (). Therefore, a set of variables P () satisfing the Kuhn Tucker condition is sufficient for maimization of the mutual information under the constraints P () = 1 and P () 0. Based on (7) for a pair of random variables X; Y, in [10, Theorem 4.5.1] it is shown that the Kuhn Tucker condition is equivalent to having a constant C such that I(; Y )=C; P () > 0 I(; Y ) C; P () =0: (12) This specialization of the Kuhn Tucker condition suggests that an observer Y can get maimum average information about X if each for which P () > 0 has the same information for the observer. From (3), the unique number C satisfing (12) is the maimum of the mutual information I(X; Y ). The condition (12) ielded a criterion for testing (and eventuall to an iterative algorithm for finding) the capacit-achieving distribution for single-user channels. III. TOTAL CAPACITY An -user discrete memorless AC is defined b input alphabets X m, m =1; 2;...;, an output alphabet Y, and a conditional probabilit distribution Q(Y j X). Without loss of generalit, assume that for each 2Ythere eists a set of values ( 1 ; 2 ;...; ), for which Q( j 1; 2;...; ) 6= 0.Aregular channel is a AC with the condition jx m j jyj for all m. The total capacit of a AC is the solution of (1), where the joint distribution is given b P 1(X 1)P 2(X 2)...P (X ) Q(Y j X). Although the objective function is concave in the probabilit distribution for an one particular user, it is not concave over variation across users [6]. Despite this nonconveit, it has been shown in [8] that for regular channels the Kuhn Tucker condition is sufficient for a distribution to achieve the total capacit. Therefore, the total capacit of a regular channel can be obtained b finding an input distribution that satisfies the Kuhn Tucker condition. Nonregular channels can be turned into regular channels via reduction of the input alphabets. Each regular channel resulting from deletion of particular letters from the input alphabets is called a sub-ac. It has been shown in [8] that there eists a sub-ac with capacit equal to the original channel. Therefore, the capacit of a AC is the maimum of the capacit of all possible sub-acs. Note that reduction of the input alphabets cannot increase capacit. Since each sub-ac is regular, its capacit can be obtained b finding a distribution that satisfies the Kuhn Tucker condition. Accordingl, the discussion in this paper is restricted to regular channels. The following theorem due to [8] states the necessar and sufficient 1 condition for a capacit-achieving distribution in a regular channel. To emphasize the similarit to the single-user case (12), the theorem has been reformulated in terms of the information function I m ( m ; Y ) defined in the previous section. Theorem 1: For a regular AC, a set of necessar and sufficient conditions for the input probabilit distributions P m (X m ), m = 1; 2;...;, to achieve the total capacit is that there eists a number C such that for all m and m I m ( m ; Y )=C; P m ( m ) > 0 I m(m; Y ) C; P m(m) =0: (13) Furthermore, the total capacit of the channel is C total = C. Proof: Application of the Kuhn Tucker condition for the maimization of I(X; Y ) subject to the constraints P m(m) =1, and P m ( m ) 0, m 2, shows that if the input distribution P 3 () = m=1 P 3 m( m) achieves the total capacit then there eists unique Lagrange multiplier vectors 3 =( 1 ; 2 ;...; ) and 3 =( 1;1 ; 1;2 ;...; 1;jX j ; 2;1 ;...; ;jx j ) such that for all m and Y m(m) + m + m; =0 m; =0; Pm(m) > 0 m; 0; P m ( m )=0: In other words, there eist numbers m such that for all m and Y ) = 0 m m(m) P m ( m ) > Y ) 0 m m(m) P m ( m )=0: Since the channel is regular, according to [8] the above condition is also a sufficient condition for the global optimalit of the distributions P m (X m ). 1 Full details of the sufficienc proof in [8] can be found in [9].

3 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER Using (9), it follows that the set of distributions P m(xm) achieves the total capacit if and onl if for all m and m From (5) for an m, i.e., I m ( m ; Y )=C m ; P m ( m ) > 0 I m ( m ; Y ) C m ; P m ( m )=0: (14) P m ( m )I m ( m ; Y )=I(X; Y ) P m(m)cm = Cm is invariant with m and is equal to the maimum of mutual information. Therefore, in (14) C m = C = Ctotal for all m. Theorem 1 is a generalization of (12) for maimization of I(X; Y ). It shows that an observer Y with sufficient domain of observation can get maimum average information about a set of independent sources X 1 ;X 2 ;...;X, if each probable outcome of each source gives the same average (over all possible outcomes of all other sources) information. An obvious avenue of generalization is to consider maimization of the conditional mutual informations, I(XS ; Y jxs ), to obtain bounds on R(S) i2s R i, S. Let I(; Y jz) =D (P (Y j ; z)kp (Y jz)) I(; Y jz) = z P (z)i(; Y jz): Then for a regular AC, a set of necessar, but not sufficient conditions for the input distributions P m (X m ) to maimize R(S) (over the space of input product distributions), is that there eists a number C such that for all m 2Sand m 0 2 S I m(m; Y jxs )=C; P m(m) > 0 I m ( m ; Y jxs ) C; P m( m )=0 I(XS; Y jxsnfm g; m )=C; P m ( m ) > 0 I(XS; Y jxsnfm g; m ) C; P m ( m )=0: In the case of a true maimum, ma R(S) =C. The nonsufficienc results from the fact that although the Kuhn Tucker conditions are necessar and sufficient for maimization of each I(XS; Y jxs = S ), this does not impl sufficienc for a conve combination P (S )I(X S; Y jxs = S ). Theorem 1 cannot be used directl to obtain an analtical solution for maimization of I(X; Y ). It can however be used to obtain an iterative solution for the total capacit of the AC. IV. ITERATIVE COPUTATION OF THE TOTAL CAPACITY In this section, a generalization of the Arimoto Blahut algorithm [1], [2] for computation of the total capacit of regular AC is presented. The Arimoto Blahut algorithm gives a sequence of input probabilit distributions P r (), r = 0; 1;... that converges to a capacit-achieving distribution for a single-user channel. This sequence is defined b P r+1 () =P r () for all 2X, where P 0 () 6= 0for all. ep(i(; Y )) P r ( 0 ) ep(i( 0 ; Y )) (15) The generalization of the Arimoto Blahut algorithm for computation of the total capacit of regular ACs is given b the following sequence of probabilit distributions in r = 0; 1;... for all m = 1; 2...; and m 2X m: P r+1 m ( m )=P r m( m ) ep(i m(m; Y )) P r m( 0 m) ep(i m( 0 m ; Y )) (16) where P 0 m( m ) 6= 0, for all m and m. The probabilit distribution for user m is calculated based on the updated probabilit distributions for users 1 to m01.for =1, this reduces to the usual Arimoto Blahut algorithm (15). Theorem 2: For a regular AC, the sequence (16) converges to a total capacit-achieving distribution. The theorem is established b showing that the sequence (16) converges, and that an convergence point satisfies (13). Since mutual information is finite, the convergence of the sequence can be verified b showing that the sequence is nondecreasing in mutual information, which will be shown in Lemma 1. The characterization of the convergence point of the sequence is given in Lemma 2, which is actuall applicable to a wider class of algorithms, where ep in (16) can be replaced b an arbitrar monotonic function. Thus, Lemmas 1 and 2 establish Theorem 2. Lemma 1: The sequence (16) is nondecreasing in mutual information. Proof: The sequence (16) has two nested loops. The outer loop is over m, i.e., different users. For each user, the inner loop is over the letters m 2X m. It suffices to show that for each user m, updating the probabilities according to (16), increases mutual information. For fied m let and P () = P 0 () = P n ( n ) P 0 n( n)=p 0 m ( m) P n(n) be two input product distributions with the same marginals for all users, ecept possibl for user m. Given the channel transition matri Q( j ), define S(;P 0 )= Q( j )log Sm( 0 m ;P 0 )= J(P; P 0 )= S(;P 0 ) S(;P 0 ) Q( j ) Pn( 0 n) Q( j 0 ) P 0 n( 0 n) (17) P 0 n( n ) (18) P n ( n )+ H(X n ) (19) n n where H(X n) is with Pn. Note J(P; P ) = I(X; Y ) where the mutual information is with the distribution P. oreover Sm( 0 0 m;p ) = log P (m) +Im(m; Y ) 0 H(X n) (20) where I m ( m ; Y ) is with the distribution P. For fied P 0, J(P; P 0 ) is a concave function of P. This is because H(X m) is a concave function of Pm, which is a marginal of P, and the first element in the right-hand side of (19) is a linear function of P m. Therefore, the solution to r P J(P; P 0 )=0maimizes J(P; P 0 ) for fied P 0.

4 2782 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER 2004 Consider the constraint P m( 3 m)=10 P m( m)=1b substituting 6= P m( m) where 3 is an selected element of X m. Then P m is a function of jx m j01 variables, P 0 = S0 m( m ;P 0 ) 0 Sm( 0 3 m;p 0 )+log Pm( 3 m ) m( m) P m( m) for m 6= 3 m. Therefore, the solution to r P Thus, J(P m ;P 0 )=0is given b log P m ( m ) 0 S 0 m( m ;P 0 )=c m ; for all m 6= 3 m: P m( m)= for all m 6= 3 m and from (20) P m( m)= which reduces to P m( m)=p 0 m( m) ep(s 0 m( m ;P 0 )) ep(s 0 m( 0 m;p 0 )) Pm( 0 m ) ep I m ( m ; Y ) 0 H(X n ) Pm( 0 0 m) ep I m( 0 m; Y ) 0 H(X n) ep(i m( m; Y )) Pm( 0 0 m): ep(i m( 0 m; Y )) : (21) Therefore, for fied P 0, the function J(P; P 0 ) is maimized for P m selected according to (21). Denoting the distribution P with this P m as P 3 I(P 0 )=J(P 0 ;P 0 ) J(P 3 ;P 0 ): (22) On the other hand, using a similar inequalit to that used in [2] J(P; P 0 ) J(P; P )=I(P ) (23) where P 0 m 6= P m. Now the two inequalities (22) and (23) show that updating distribution of the user m from P 0 to P 3 according to (21) does not decrease mutual information. Now define the sequence of input probabilit distributions P r m( m ), r =0; 1;... P r+1 m ( m )=P r m( m ) f (I m( m; Y )) P r m( 0 m)f(i m( 0 m; Y )) (24) for all m =1; 2;...; and m 2X m, where f is an continuous monotonicall increasing positive function over R +. Also, assume that I m ( m ; Y ) is calculated similar to (16) and that P 0 m( m ) 6= 0, for all m and m. The sequence (24) is defined to emphasize that the ke propert required for the following lemma is monotonicit, rather than special properties of the eponent in (16). Lemma 2: For a regular AC, the convergence point of the sequence P r in (24), if it eists, achieves the total capacit. Proof: Suppose that for each m and m, the sequence P r m( m), r!1converges to P 3 m( m ). It will be shown that the set of probabilities P 3 m( m ) satisfies both of the conditions of (13). From the sequence P r m( m) define the sequence D r m( m )= f (I m( m; Y )) P r m( 0 m)f(i m( 0 m; Y )) : Convergence of P r m( m) implies that lim r!1 D r m ( m)=1for all m and m such that P 3 m( m ) 6= 0. The reason is that for r!1 Pm r+1 ( m)=pm( r m)=pm( 3 m): Denoting I 3 m( m ; Y ) as I m ( m ; Y ) for the distribution P 3 f (I 3 m( m ; Y )) = P 3 m( 0 m)f(i 3 m( 0 m; Y )) subject to P 3 m( m) 6= 0, which in turn implies f (I 3 m( m; Y )) = C m, for all m such that P 3 m( m ) 6= 0, for some constant C m. Since f is monotonic, this implies the first condition of (13). The net step is to show that P 3 not satisfing the second condition of (13) contradicts the convergence of P r! P 3.Now P r m( m)=p 0 m( m) r D n m( m): (25) If P 3 does not satisf the second condition of (13), then since f is assumed to be monotonicall increasing f (I 3 m( m ; Y )) > P 3 m( 0 m)f(i 3 m( 0 m; Y )) for some m, and m such that P 3 m( m)=0. This means that the sequence D r m( m ) for some m and m has converged to a value greater than one. This, along with the assumption P 0 m( m ) > 0, contradicts the convergence of the sequence of P r m( m) as a sequence of partial products in (25). As the proof shows, the requirement that the initial probabilities to be nonzero for all letters of all users is essential for convergence to a Kuhn Tucker point. A starting zero probabilit for a letter remains alwas zero. Theorem 2 is now established b Lemmas 1 and 2. From Theorem 1 and Lemma 2 it can be seen that if in (24) for a function f and a number m, and each value of m, the variation of I m ( m ; Y ) in the sequence P r tends to zero, then the total capacit of the channel is C total = I m( m; Y ), for m such that P ( m) 6= 0. Therefore, in a generalized algorithm based on (24), the convergence of the algorithm can be considered as a stopping criterion, or nonconvergence after a certain number of iteration as an initialization. Under general selection of f in Lemma 2, the algorithm converges onl to the optimal probabilit distribution (if it converges). Arbitrar selection of f does not guarantee that the algorithm will alwas converge. Such a guarantee required special properties of the eponential function in Lemma 1. Numerical investigations show, however, that the sequence almost alwas converges for a wide range of choices of the function f, and that it tpicall converges faster than f = ep. Fig. 1 shows tpical results for a two-user channel with X 1 = X 2 = f0; 1g, Y = f0; 1; 2g, and transition matri 0:2 0:3 0:5 0:7 0:2 0:1 0:5 0:1 0:4 0:3 0:4 0:3 (rows are inputs, naturall ordered). The initial distributions P (X 1 = 0) = 0:3 and P (X 2 =0)=0:6 were used in the eample. The figure shows the sequence (16) for various choices of f, namel, f () =e, f () = p, f () =, and f () = e 0 1 (the latter two result in almost identical sequences). The results of Fig. 1 seem to be representative, in that similar behavior was observed, in terms of relative convergence speed for man other (albeit arbitraril chosen) channels (with differing input/output alphabet sizes).

5 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER Fig. 1. Tpical algorithm convergence for different functions f (). Fig. 2. Values of (p ;p ) resulting in optimalit of the uniform distribution for the channel (27). An eplanation for convergence in (24) is as follows. At each iteration, the algorithm updates the probabilit distributions b giving more probabilit to letters that have more value of the information function. Increasing the probabilit for high information letters has two effects. First, it increases the average information I(X; Y ). On the other hand, this increased probabilit decreases the information of those letters, thus balancing the information function toward a constant, a requirement for the maimizing distribution. Conversel, the reduction in probabilit of low information letters is in favor of increasing information of those letters. This acts toward the balance of information between letters unless the probabilit of such low information alphabets reaches zero, i.e., the second condition of (13) is satisfied. An interesting open problem is to find a better characterization of monotonic functions f in (24) such that convergence alwas occurs, i.e., find an analog of Lemma 1 for functions other than ep. The numerical results provide strong motivation for further investigation in that direction. V. SYETRIC CHANNELS A single-user discrete memorless channel (DC) is defined b an input alphabet X, an output alphabet Y, and a conditional probabilit distribution Q(Y j X). The Kuhn Tucker condition [10, Theorem 4.5.1] implies that an input distribution P (X) is optimal if there eists a constant C, such that (12) is satisfied. This shows that the set of all channels with a specific nonboundar distribution (a distribution with nonzero probabilit for all letters) P 3 () as optimal, are channels for which the information function I(; Y ) for the distribution P 3 (X) is constant for all. Of particular interest is the class of channels for which the uniform distribution U () =1= jx j is the optimal input distribution. For such channels, the function Q( j ) T () = Q( j )log 2Y 2X Q( j 0 ) = I(; Y ) 0 log jx j (26) needs to be constant. The term I(; Y ) is the information function for the uniform distribution. For the class of smmetric channels identified in Shannon s original work [11, Sec. 16] and further developed in [10, Sec. 4.5], it is known that the uniform distribution is optimal. The proof of this result, however, rests onl on the fact that for such channels I(; Y ) is constant. This propert is equivalent to satisfaction of the Kuhn Tucker condition for uniform distribution. Thus, the whole class of channels with optimal uniform distribution is identified b uniformit in T () instead of permutation invariance in the channel transition probabilit table Q(Y jx). Note that T () is a function of this table. It could therefore be more natural to define a concept of smmetr b uniformit in T (). This propert for a channel is equivalent to optimalit of uniform input distribution. The purpose of this section is to etend this new definition of smmetr to the AC. For the purpose of distinction with the previous notions of smmetr, this definition will be referred to as T -smmetr. Definition 1 (T -Smmetric DC): A DC is T -smmetric if the function T defined b (26) is constant. Channels with row and column permutation smmetr in the transition probabilit matri (i.e., smmetric according to [10, p. 94]) are T -smmetric and it is well known that the are optimized b uniform input distribution. It is possible however to construct nontrivial eamples of T -smmetric channels that are not smmetric according to [10] (even allowing for partitions of the matri). One such eample is the channel defined b the conditional probabilit table 1=3 1=3 1=3 (27) p 1 p p 1 0 p 2 with p 1 = p 2 =1=6. Direct calculation shows that the uniform distribution satisfies (12) for this channel. Furthermore, this channel is not some isolated special case. In fact, infinitel man channels ma be constructed with constant T () that are not smmetric according to [10]. In demonstration, Fig. 2 shows the locus in the p 1 ;p 2 plane of all channels of the form (27) having uniform input distribution as optimal. Theorem 1 shows that the set of all regular ACs with a set of nonboundar distributions Pm( 3 m ) as optimal distribution, are channels for which I m(m; Y ) for that distribution is a constant. Of special interest is the class of ACs for which the uniform input distribution for each user is optimal. The T -smmetric single-user channel in Definition 1 is a special case of this class for single-user channels. The class of T -smmetric ACs will be defined based on a function of the channel conditional probabilities. For a AC Q( j 1; 2; ::; ), define the function T () and its marginalization T m ( m ) to user m b T () = T m ( m )= Q( j ) log Q( j ) Q( j 0 ) (28) T (): (29) Definition 2 (T -Smmetric AC): A AC is T -smmetric if for all m, the functions T m defined b (29) are constant.

6 2784 IEEE TRANSACTIONS ON INFORATION THEORY, VOL. 50, NO. 11, NOVEBER 2004 It can be inferred that for T -smmetric channels the dependenc of T m ( m ) on m can onl be in the form of T m ( m )=C= jx m j, where C is a constant. Using Theorem 1 one can show that the T -smmetr propert for a regular channel is equivalent to the propert that the uniform distribution for all users is the capacit-achieving distribution. In fact, for the uniform input distribution which leads to T m(a) = = T (; m = a) T ( a;b (; m = a)) and I(; Y )=T () + I m ( m ; Y )= = l6=m 1 jx l j l6=m 1 jx l j m log(jx m j) I(; Y ) T m ( m )+ m log(jx m j): (30) = T (; m = b) = T m(b) which means the channel is T -smmetric. One eample of a smmetric AC is the on off fading AC defined in [12]. This channel is a regular AC, and it can be shown that in the special case of unique outputs [12], the channel is smmetric according to Definition 3. Therefore, the uniform distribution achieves capacit for this channel. If T m ( m ) is independent of m, then so is I m ( m ; Y )=C m. Since I(X; Y )= I m ( m ; Y )= jx m j = C m is independent of m, and I m ( m ; Y )=C, which implies that T -smmetr results in (13) for uniform distributions. Conversel, if for the uniform input distribution I m(m; Y )=C, then from (30), Tm(m) can onl be dependent on m. Equation (30) shows that the total capacit of a regular T -smmetric channel is C total = Tm(m) l6=m jx lj + l log(jx l j) (31) for an arbitrar m and m. Finall, a subset of T -smmetric channels can be identified according to a certain smmetr propert of their transition probabilities. Definition 3 (Smmetric AC): A AC with identical 2 input alphabets X is smmetric if each row permutation of the AC transition probabilit matri corresponding to transposing source letters a; b 2X for all users is the same as a column permutation. The specialization of the condition in Definition 3 to single-user channels gives a subset of the condition in [10] for a smmetric singleuser channel. Theorem 3: The uniform input distribution achieves the total capacit of a regular smmetric AC. Proof: According to the preceding discussion, it suffices to show that an smmetric AC is T -smmetric. Define the operation a;b (i 1;i 2;...;i ) that swaps a and b in its argument, e.g., For the smmetric AC 2;3(1; 3; 4; 2; 3) = (1; 2; 4; 3; 2): Q( j a;b ()) T ( a;b ())= Q( j a;b ()) log Q( j a;b ()) 0 0 Q( j ) = Q( j ) log Q(0 j ) = T () VI. CONCLUSION The Arimoto Blahut algorithm has been etended for computation of the total capacit of regular ACs. The total capacit of the general ACs can be broken down to capacit computation for a number of regular ACs. oreover, a class of regular T -smmetric discrete memorless ACs has been defined, with the propert that the uniform input distribution for each user achieves the total capacit. Specialization of this class to the single-user case gives an etended concept of smmetr, namel, T -smmetr, which is the class of all DCs with uniform optimum input distribution. A subclass of ACs was also identified in which smmetr is a direct result of a transposition smmetr in the channel transition probabilit. REFERENCES [1] S. Arimoto, An algorithm for computing the capacit of arbitrar discrete memorless channels, IEEE Trans. Inform. Theor, vol. IT-18, pp , Jan [2] R. E. Blahut, Computation of channel capacit and rate-distortion functions, IEEE Trans. Inform. Theor, vol. IT-18, pp , Jul [3] R. W. Yeung, A First Course in Information Theor. New York: Kluwer Academic/Plenum, [4]. Chiang and S. Bod, Geometric programming duals of channel capacit and rate distortion, IEEE Trans. Inform. Theor, vol. 50, pp , Feb [5] J. Huang and S. en, Characterization and computation of optimal distributions for channel coding, in Proc. 37th Conf. Information Science and Sstems, Baltimore, D, ar [6] Y. Watanabe, The total capacit of two-user multiple-access channel with binar output, IEEE Trans. Inform. Theor, vol. 42, pp , Sept [7] S. Bod and L. Vandenberghe, Conve Optimization. Cambridge, U.K.: Cambridge Univ. Press, [8] Y. Watanabe and K. Kamoi, The total capacit of multiple access channel, in Proc. IEEE Int. Smp. Information Theor, Lausanne, Switzerland, June/Jul 2002, p [9], The total capacit of multiple access channel, IEEE Trans. Inform. Theor, submitted for publication. [10] R. G. Gallager, Information Theorand Reliable Communication. New York: Wile, [11] C. E. Shannon, A mathematical theor of communication, Bell Sst. Tech. J., vol. 27, pp , [12] E. Perron,. Rezaeian, and A. Grant, The on-off fading channel, in Proc. IEEE Int. Smp. Information Theor, Yokohama, Japan, June/Jul 2003, p Obviousl, the equalit of alphabet sets is not important. It is sufficient that the alphabet sets can be mapped to an identical set.

Computation of Csiszár s Mutual Information of Order α

Computation of Csiszár s Mutual Information of Order α Computation of Csiszár s Mutual Information of Order Damianos Karakos, Sanjeev Khudanpur and Care E. Priebe Department of Electrical and Computer Engineering and Center for Language and Speech Processing

More information

Stochastic Interpretation for the Arimoto Algorithm

Stochastic Interpretation for the Arimoto Algorithm Stochastic Interpretation for the Arimoto Algorithm Serge Tridenski EE - Sstems Department Tel Aviv Universit Tel Aviv, Israel Email: sergetr@post.tau.ac.il Ram Zamir EE - Sstems Department Tel Aviv Universit

More information

School of Computer and Communication Sciences. Information Theory and Coding Notes on Random Coding December 12, 2003.

School of Computer and Communication Sciences. Information Theory and Coding Notes on Random Coding December 12, 2003. ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE School of Computer and Communication Sciences Handout 8 Information Theor and Coding Notes on Random Coding December 2, 2003 Random Coding In this note we prove

More information

In applications, we encounter many constrained optimization problems. Examples Basis pursuit: exact sparse recovery problem

In applications, we encounter many constrained optimization problems. Examples Basis pursuit: exact sparse recovery problem 1 Conve Analsis Main references: Vandenberghe UCLA): EECS236C - Optimiation methods for large scale sstems, http://www.seas.ucla.edu/ vandenbe/ee236c.html Parikh and Bod, Proimal algorithms, slides and

More information

Flows and Connectivity

Flows and Connectivity Chapter 4 Flows and Connectivit 4. Network Flows Capacit Networks A network N is a connected weighted loopless graph (G,w) with two specified vertices and, called the source and the sink, respectivel,

More information

6. Vector Random Variables

6. Vector Random Variables 6. Vector Random Variables In the previous chapter we presented methods for dealing with two random variables. In this chapter we etend these methods to the case of n random variables in the following

More information

An Information Theory For Preferences

An Information Theory For Preferences An Information Theor For Preferences Ali E. Abbas Department of Management Science and Engineering, Stanford Universit, Stanford, Ca, 94305 Abstract. Recent literature in the last Maimum Entrop workshop

More information

The Entropy Power Inequality and Mrs. Gerber s Lemma for Groups of order 2 n

The Entropy Power Inequality and Mrs. Gerber s Lemma for Groups of order 2 n The Entrop Power Inequalit and Mrs. Gerber s Lemma for Groups of order 2 n Varun Jog EECS, UC Berkele Berkele, CA-94720 Email: varunjog@eecs.berkele.edu Venkat Anantharam EECS, UC Berkele Berkele, CA-94720

More information

Linear programming: Theory

Linear programming: Theory Division of the Humanities and Social Sciences Ec 181 KC Border Convex Analsis and Economic Theor Winter 2018 Topic 28: Linear programming: Theor 28.1 The saddlepoint theorem for linear programming The

More information

CPS 616 ITERATIVE IMPROVEMENTS 10-1

CPS 616 ITERATIVE IMPROVEMENTS 10-1 CPS 66 ITERATIVE IMPROVEMENTS 0 - APPROACH Algorithm design technique for solving optimization problems Start with a feasible solution Repeat the following step until no improvement can be found: change

More information

On Maximizing the Second Smallest Eigenvalue of a State-dependent Graph Laplacian

On Maximizing the Second Smallest Eigenvalue of a State-dependent Graph Laplacian 25 American Control Conference June 8-, 25. Portland, OR, USA WeA3.6 On Maimizing the Second Smallest Eigenvalue of a State-dependent Graph Laplacian Yoonsoo Kim and Mehran Mesbahi Abstract We consider

More information

ANALYTIC CENTER CUTTING PLANE METHODS FOR VARIATIONAL INEQUALITIES OVER CONVEX BODIES

ANALYTIC CENTER CUTTING PLANE METHODS FOR VARIATIONAL INEQUALITIES OVER CONVEX BODIES ANALYI ENER UING PLANE MEHODS OR VARIAIONAL INEQUALIIES OVER ONVE BODIES Renin Zen School of Mathematical Sciences honqin Normal Universit honqin hina ABSRA An analtic center cuttin plane method is an

More information

Perturbation Theory for Variational Inference

Perturbation Theory for Variational Inference Perturbation heor for Variational Inference Manfred Opper U Berlin Marco Fraccaro echnical Universit of Denmark Ulrich Paquet Apple Ale Susemihl U Berlin Ole Winther echnical Universit of Denmark Abstract

More information

Sample-based Optimal Transport and Barycenter Problems

Sample-based Optimal Transport and Barycenter Problems Sample-based Optimal Transport and Barcenter Problems MAX UANG New York Universit, Courant Institute of Mathematical Sciences AND ESTEBAN G. TABA New York Universit, Courant Institute of Mathematical Sciences

More information

CONTINUOUS SPATIAL DATA ANALYSIS

CONTINUOUS SPATIAL DATA ANALYSIS CONTINUOUS SPATIAL DATA ANALSIS 1. Overview of Spatial Stochastic Processes The ke difference between continuous spatial data and point patterns is that there is now assumed to be a meaningful value, s

More information

EQUIVALENT FORMULATIONS OF HYPERCONTRACTIVITY USING INFORMATION MEASURES

EQUIVALENT FORMULATIONS OF HYPERCONTRACTIVITY USING INFORMATION MEASURES 1 EQUIVALENT FORMULATIONS OF HYPERCONTRACTIVITY USING INFORMATION MEASURES CHANDRA NAIR 1 1 Dept. of Information Engineering, The Chinese Universit of Hong Kong Abstract We derive alternate characterizations

More information

Hamiltonicity and Fault Tolerance

Hamiltonicity and Fault Tolerance Hamiltonicit and Fault Tolerance in the k-ar n-cube B Clifford R. Haithcock Portland State Universit Department of Mathematics and Statistics 006 In partial fulfillment of the requirements of the degree

More information

Chapter 11 Optimization with Equality Constraints

Chapter 11 Optimization with Equality Constraints Ch. - Optimization with Equalit Constraints Chapter Optimization with Equalit Constraints Albert William Tucker 95-995 arold William Kuhn 95 oseph-ouis Giuseppe odovico comte de arane 76-. General roblem

More information

Symmetric Characterization of Finite State Markov Channels

Symmetric Characterization of Finite State Markov Channels Symmetric Characterization of Finite State Markov Channels Mohammad Rezaeian Department of Electrical and Electronic Eng. The University of Melbourne Victoria, 31, Australia Email: rezaeian@unimelb.edu.au

More information

Upper Bounds on the Capacity of Binary Intermittent Communication

Upper Bounds on the Capacity of Binary Intermittent Communication Upper Bounds on the Capacity of Binary Intermittent Communication Mostafa Khoshnevisan and J. Nicholas Laneman Department of Electrical Engineering University of Notre Dame Notre Dame, Indiana 46556 Email:{mhoshne,

More information

Ordinal One-Switch Utility Functions

Ordinal One-Switch Utility Functions Ordinal One-Switch Utilit Functions Ali E. Abbas Universit of Southern California, Los Angeles, California 90089, aliabbas@usc.edu David E. Bell Harvard Business School, Boston, Massachusetts 0163, dbell@hbs.edu

More information

Glossary. Also available at BigIdeasMath.com: multi-language glossary vocabulary flash cards. An equation that contains an absolute value expression

Glossary. Also available at BigIdeasMath.com: multi-language glossary vocabulary flash cards. An equation that contains an absolute value expression Glossar This student friendl glossar is designed to be a reference for ke vocabular, properties, and mathematical terms. Several of the entries include a short eample to aid our understanding of important

More information

Finding Limits Graphically and Numerically. An Introduction to Limits

Finding Limits Graphically and Numerically. An Introduction to Limits 60_00.qd //0 :05 PM Page 8 8 CHAPTER Limits and Their Properties Section. Finding Limits Graphicall and Numericall Estimate a it using a numerical or graphical approach. Learn different was that a it can

More information

NUMERICAL COMPUTATION OF THE CAPACITY OF CONTINUOUS MEMORYLESS CHANNELS

NUMERICAL COMPUTATION OF THE CAPACITY OF CONTINUOUS MEMORYLESS CHANNELS NUMERICAL COMPUTATION OF THE CAPACITY OF CONTINUOUS MEMORYLESS CHANNELS Justin Dauwels Dept. of Information Technology and Electrical Engineering ETH, CH-8092 Zürich, Switzerland dauwels@isi.ee.ethz.ch

More information

Communication Limits with Low Precision Analog-to-Digital Conversion at the Receiver

Communication Limits with Low Precision Analog-to-Digital Conversion at the Receiver This full tet paper was peer reviewed at the direction of IEEE Communications Society subject matter eperts for publication in the ICC 7 proceedings. Communication Limits with Low Precision Analog-to-Digital

More information

Finding Limits Graphically and Numerically. An Introduction to Limits

Finding Limits Graphically and Numerically. An Introduction to Limits 8 CHAPTER Limits and Their Properties Section Finding Limits Graphicall and Numericall Estimate a it using a numerical or graphical approach Learn different was that a it can fail to eist Stud and use

More information

On the Spectral Theory of Operator Pencils in a Hilbert Space

On the Spectral Theory of Operator Pencils in a Hilbert Space Journal of Nonlinear Mathematical Phsics ISSN: 1402-9251 Print 1776-0852 Online Journal homepage: http://www.tandfonline.com/loi/tnmp20 On the Spectral Theor of Operator Pencils in a Hilbert Space Roman

More information

UNIVERSIDAD CARLOS III DE MADRID MATHEMATICS II EXERCISES (SOLUTIONS )

UNIVERSIDAD CARLOS III DE MADRID MATHEMATICS II EXERCISES (SOLUTIONS ) UNIVERSIDAD CARLOS III DE MADRID MATHEMATICS II EXERCISES (SOLUTIONS ) CHAPTER : Limits and continuit of functions in R n. -. Sketch the following subsets of R. Sketch their boundar and the interior. Stud

More information

1.3 LIMITS AT INFINITY; END BEHAVIOR OF A FUNCTION

1.3 LIMITS AT INFINITY; END BEHAVIOR OF A FUNCTION . Limits at Infinit; End Behavior of a Function 89. LIMITS AT INFINITY; END BEHAVIOR OF A FUNCTION Up to now we have been concerned with its that describe the behavior of a function f) as approaches some

More information

On MIMO Fading Channels with Side Information at the Transmitter

On MIMO Fading Channels with Side Information at the Transmitter On MIMO Fading Channels with ide Information at the Transmitter EE850 Project, Ma 005 Tairan Wang twang@ece.umn.edu, ID: 340947 Abstract Transmission of information over a MIMO discrete-time flat fading

More information

Capacity Upper Bounds for the Deletion Channel

Capacity Upper Bounds for the Deletion Channel Capacity Upper Bounds for the Deletion Channel Suhas Diggavi, Michael Mitzenmacher, and Henry D. Pfister School of Computer and Communication Sciences, EPFL, Lausanne, Switzerland Email: suhas.diggavi@epfl.ch

More information

Journal of Inequalities in Pure and Applied Mathematics

Journal of Inequalities in Pure and Applied Mathematics Journal of Inequalities in Pure and Applied Mathematics http://jipam.vu.edu.au/ Volume 2, Issue 2, Article 15, 21 ON SOME FUNDAMENTAL INTEGRAL INEQUALITIES AND THEIR DISCRETE ANALOGUES B.G. PACHPATTE DEPARTMENT

More information

2.5. Infinite Limits and Vertical Asymptotes. Infinite Limits

2.5. Infinite Limits and Vertical Asymptotes. Infinite Limits . Infinite Limits and Vertical Asmptotes. Infinite Limits and Vertical Asmptotes In this section we etend the concept of it to infinite its, which are not its as before, but rather an entirel new use of

More information

On the relation between the relative earth mover distance and the variation distance (an exposition)

On the relation between the relative earth mover distance and the variation distance (an exposition) On the relation between the relative earth mover distance and the variation distance (an eposition) Oded Goldreich Dana Ron Februar 9, 2016 Summar. In this note we present a proof that the variation distance

More information

Additional Material On Recursive Sequences

Additional Material On Recursive Sequences Penn State Altoona MATH 141 Additional Material On Recursive Sequences 1. Graphical Analsis Cobweb Diagrams Consider a generic recursive sequence { an+1 = f(a n ), n = 1,, 3,..., = Given initial value.

More information

Chapter 6* The PPSZ Algorithm

Chapter 6* The PPSZ Algorithm Chapter 6* The PPSZ Algorithm In this chapter we present an improvement of the pp algorithm. As pp is called after its inventors, Paturi, Pudlak, and Zane [PPZ99], the improved version, pps, is called

More information

4 Inverse function theorem

4 Inverse function theorem Tel Aviv Universit, 2013/14 Analsis-III,IV 53 4 Inverse function theorem 4a What is the problem................ 53 4b Simple observations before the theorem..... 54 4c The theorem.....................

More information

Duality, Geometry, and Support Vector Regression

Duality, Geometry, and Support Vector Regression ualit, Geometr, and Support Vector Regression Jinbo Bi and Kristin P. Bennett epartment of Mathematical Sciences Rensselaer Poltechnic Institute Tro, NY 80 bij@rpi.edu, bennek@rpi.edu Abstract We develop

More information

Research Article Equivalent Elastic Modulus of Asymmetrical Honeycomb

Research Article Equivalent Elastic Modulus of Asymmetrical Honeycomb International Scholarl Research Network ISRN Mechanical Engineering Volume, Article ID 57, pages doi:.5//57 Research Article Equivalent Elastic Modulus of Asmmetrical Honecomb Dai-Heng Chen and Kenichi

More information

Optimal scaling of the random walk Metropolis on elliptically symmetric unimodal targets

Optimal scaling of the random walk Metropolis on elliptically symmetric unimodal targets Optimal scaling of the random walk Metropolis on ellipticall smmetric unimodal targets Chris Sherlock 1 and Gareth Roberts 2 1. Department of Mathematics and Statistics, Lancaster Universit, Lancaster,

More information

Nash Equilibrium and the Legendre Transform in Optimal Stopping Games with One Dimensional Diffusions

Nash Equilibrium and the Legendre Transform in Optimal Stopping Games with One Dimensional Diffusions Nash Equilibrium and the Legendre Transform in Optimal Stopping Games with One Dimensional Diffusions J. L. Seton This version: 9 Januar 2014 First version: 2 December 2011 Research Report No. 9, 2011,

More information

Organization of a Modern Compiler. Front-end. Middle1. Back-end DO I = 1, N DO J = 1,M S 1 1 N. Source Program

Organization of a Modern Compiler. Front-end. Middle1. Back-end DO I = 1, N DO J = 1,M S 1 1 N. Source Program Organization of a Modern Compiler ource Program Front-end snta analsis + tpe-checking + smbol table High-level ntermediate Representation (loops,arra references are preserved) Middle loop-level transformations

More information

Functions of Several Variables

Functions of Several Variables Chapter 1 Functions of Several Variables 1.1 Introduction A real valued function of n variables is a function f : R, where the domain is a subset of R n. So: for each ( 1,,..., n ) in, the value of f is

More information

CHAPTER 3 Applications of Differentiation

CHAPTER 3 Applications of Differentiation CHAPTER Applications of Differentiation Section. Etrema on an Interval................... 0 Section. Rolle s Theorem and the Mean Value Theorem...... 0 Section. Increasing and Decreasing Functions and

More information

AE/ME 339. K. M. Isaac Professor of Aerospace Engineering. December 21, 2001 topic13_grid_generation 1

AE/ME 339. K. M. Isaac Professor of Aerospace Engineering. December 21, 2001 topic13_grid_generation 1 AE/ME 339 Professor of Aerospace Engineering December 21, 2001 topic13_grid_generation 1 The basic idea behind grid generation is the creation of the transformation laws between the phsical space and the

More information

4.7. Newton s Method. Procedure for Newton s Method HISTORICAL BIOGRAPHY

4.7. Newton s Method. Procedure for Newton s Method HISTORICAL BIOGRAPHY 4. Newton s Method 99 4. Newton s Method HISTORICAL BIOGRAPHY Niels Henrik Abel (18 189) One of the basic problems of mathematics is solving equations. Using the quadratic root formula, we know how to

More information

QUADRATIC AND CONVEX MINIMAX CLASSIFICATION PROBLEMS

QUADRATIC AND CONVEX MINIMAX CLASSIFICATION PROBLEMS Journal of the Operations Research Societ of Japan 008, Vol. 51, No., 191-01 QUADRATIC AND CONVEX MINIMAX CLASSIFICATION PROBLEMS Tomonari Kitahara Shinji Mizuno Kazuhide Nakata Toko Institute of Technolog

More information

Stability Analysis for Linear Systems under State Constraints

Stability Analysis for Linear Systems under State Constraints Stabilit Analsis for Linear Sstems under State Constraints Haijun Fang Abstract This paper revisits the problem of stabilit analsis for linear sstems under state constraints New and less conservative sufficient

More information

Optimization in Information Theory

Optimization in Information Theory Optimization in Information Theory Dawei Shen November 11, 2005 Abstract This tutorial introduces the application of optimization techniques in information theory. We revisit channel capacity problem from

More information

1 Kernel methods & optimization

1 Kernel methods & optimization Machine Learning Class Notes 9-26-13 Prof. David Sontag 1 Kernel methods & optimization One eample of a kernel that is frequently used in practice and which allows for highly non-linear discriminant functions

More information

Steepest descent on factor graphs

Steepest descent on factor graphs Steepest descent on factor graphs Justin Dauwels, Sascha Korl, and Hans-Andrea Loeliger Abstract We show how steepest descent can be used as a tool for estimation on factor graphs. From our eposition,

More information

CSE 546 Midterm Exam, Fall 2014

CSE 546 Midterm Exam, Fall 2014 CSE 546 Midterm Eam, Fall 2014 1. Personal info: Name: UW NetID: Student ID: 2. There should be 14 numbered pages in this eam (including this cover sheet). 3. You can use an material ou brought: an book,

More information

NATIONAL UNIVERSITY OF SINGAPORE Department of Mathematics MA4247 Complex Analysis II Lecture Notes Part I

NATIONAL UNIVERSITY OF SINGAPORE Department of Mathematics MA4247 Complex Analysis II Lecture Notes Part I NATIONAL UNIVERSITY OF SINGAPORE Department of Mathematics MA4247 Comple Analsis II Lecture Notes Part I Chapter 1 Preliminar results/review of Comple Analsis I These are more detailed notes for the results

More information

Computation of Information Rates from Finite-State Source/Channel Models

Computation of Information Rates from Finite-State Source/Channel Models Allerton 2002 Computation of Information Rates from Finite-State Source/Channel Models Dieter Arnold arnold@isi.ee.ethz.ch Hans-Andrea Loeliger loeliger@isi.ee.ethz.ch Pascal O. Vontobel vontobel@isi.ee.ethz.ch

More information

Math 20 Spring 2005 Final Exam Practice Problems (Set 2)

Math 20 Spring 2005 Final Exam Practice Problems (Set 2) Math 2 Spring 2 Final Eam Practice Problems (Set 2) 1. Find the etreme values of f(, ) = 2 2 + 3 2 4 on the region {(, ) 2 + 2 16}. 2. Allocation of Funds: A new editor has been allotted $6, to spend on

More information

Discrete Memoryless Channels with Memoryless Output Sequences

Discrete Memoryless Channels with Memoryless Output Sequences Discrete Memoryless Channels with Memoryless utput Sequences Marcelo S Pinho Department of Electronic Engineering Instituto Tecnologico de Aeronautica Sao Jose dos Campos, SP 12228-900, Brazil Email: mpinho@ieeeorg

More information

Language and Statistics II

Language and Statistics II Language and Statistics II Lecture 19: EM for Models of Structure Noah Smith Epectation-Maimization E step: i,, q i # p r $ t = p r i % ' $ t i, p r $ t i,' soft assignment or voting M step: r t +1 # argma

More information

Section 1.5 Formal definitions of limits

Section 1.5 Formal definitions of limits Section.5 Formal definitions of limits (3/908) Overview: The definitions of the various tpes of limits in previous sections involve phrases such as arbitraril close, sufficientl close, arbitraril large,

More information

Convexity/Concavity of Renyi Entropy and α-mutual Information

Convexity/Concavity of Renyi Entropy and α-mutual Information Convexity/Concavity of Renyi Entropy and -Mutual Information Siu-Wai Ho Institute for Telecommunications Research University of South Australia Adelaide, SA 5095, Australia Email: siuwai.ho@unisa.edu.au

More information

MAT 1275: Introduction to Mathematical Analysis. Graphs and Simplest Equations for Basic Trigonometric Functions. y=sin( x) Function

MAT 1275: Introduction to Mathematical Analysis. Graphs and Simplest Equations for Basic Trigonometric Functions. y=sin( x) Function MAT 275: Introduction to Mathematical Analsis Dr. A. Rozenblum Graphs and Simplest Equations for Basic Trigonometric Functions We consider here three basic functions: sine, cosine and tangent. For them,

More information

Square Estimation by Matrix Inverse Lemma 1

Square Estimation by Matrix Inverse Lemma 1 Proof of Two Conclusions Associated Linear Minimum Mean Square Estimation b Matri Inverse Lemma 1 Jianping Zheng State e Lab of IS, Xidian Universit, Xi an, 7171, P. R. China jpzheng@idian.edu.cn Ma 6,

More information

8.4. If we let x denote the number of gallons pumped, then the price y in dollars can $ $1.70 $ $1.70 $ $1.70 $ $1.

8.4. If we let x denote the number of gallons pumped, then the price y in dollars can $ $1.70 $ $1.70 $ $1.70 $ $1. 8.4 An Introduction to Functions: Linear Functions, Applications, and Models We often describe one quantit in terms of another; for eample, the growth of a plant is related to the amount of light it receives,

More information

SHANNON made the following well-known remarks in

SHANNON made the following well-known remarks in IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 50, NO. 2, FEBRUARY 2004 245 Geometric Programming Duals of Channel Capacity and Rate Distortion Mung Chiang, Member, IEEE, and Stephen Boyd, Fellow, IEEE

More information

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures AB = BA = I,

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures AB = BA = I, FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING Lectures MODULE 7 MATRICES II Inverse of a matri Sstems of linear equations Solution of sets of linear equations elimination methods 4

More information

H(X) = plog 1 p +(1 p)log 1 1 p. With a slight abuse of notation, we denote this quantity by H(p) and refer to it as the binary entropy function.

H(X) = plog 1 p +(1 p)log 1 1 p. With a slight abuse of notation, we denote this quantity by H(p) and refer to it as the binary entropy function. LECTURE 2 Information Measures 2. ENTROPY LetXbeadiscreterandomvariableonanalphabetX drawnaccordingtotheprobability mass function (pmf) p() = P(X = ), X, denoted in short as X p(). The uncertainty about

More information

Research Article On Sharp Triangle Inequalities in Banach Spaces II

Research Article On Sharp Triangle Inequalities in Banach Spaces II Hindawi Publishing Corporation Journal of Inequalities and Applications Volume 2010, Article ID 323609, 17 pages doi:10.1155/2010/323609 Research Article On Sharp Triangle Inequalities in Banach Spaces

More information

CHAPTER 2 THE MATHEMATICS OF OPTIMIZATION. Solutions. dπ =

CHAPTER 2 THE MATHEMATICS OF OPTIMIZATION. Solutions. dπ = CHAPTER THE MATHEMATICS OF OPTIMIZATION The problems in this chapter are primaril mathematical. The are intended to give students some practice with taking derivatives and using the Lagrangian techniques,

More information

Interspecific Segregation and Phase Transition in a Lattice Ecosystem with Intraspecific Competition

Interspecific Segregation and Phase Transition in a Lattice Ecosystem with Intraspecific Competition Interspecific Segregation and Phase Transition in a Lattice Ecosstem with Intraspecific Competition K. Tainaka a, M. Kushida a, Y. Ito a and J. Yoshimura a,b,c a Department of Sstems Engineering, Shizuoka

More information

You don't have to be a mathematician to have a feel for numbers. John Forbes Nash, Jr.

You don't have to be a mathematician to have a feel for numbers. John Forbes Nash, Jr. Course Title: Real Analsis Course Code: MTH3 Course instructor: Dr. Atiq ur Rehman Class: MSc-II Course URL: www.mathcit.org/atiq/fa5-mth3 You don't have to be a mathematician to have a feel for numbers.

More information

A set C R n is convex if and only if δ C is convex if and only if. f : R n R is strictly convex if and only if f is convex and the inequality (1.

A set C R n is convex if and only if δ C is convex if and only if. f : R n R is strictly convex if and only if f is convex and the inequality (1. ONVEX OPTIMIZATION SUMMARY ANDREW TULLOH 1. Eistence Definition. For R n, define δ as 0 δ () = / (1.1) Note minimizes f over if and onl if minimizes f + δ over R n. Definition. (ii) (i) dom f = R n f()

More information

On Information and Sufficiency

On Information and Sufficiency On Information and Sufficienc Huaiu hu SFI WORKING PAPER: 997-02-04 SFI Working Papers contain accounts of scientific work of the author(s) and do not necessaril represent the views of the Santa Fe Institute.

More information

All parabolas through three non-collinear points

All parabolas through three non-collinear points ALL PARABOLAS THROUGH THREE NON-COLLINEAR POINTS 03 All parabolas through three non-collinear points STANLEY R. HUDDY and MICHAEL A. JONES If no two of three non-collinear points share the same -coordinate,

More information

Chapter 6. Nonlinear Equations. 6.1 The Problem of Nonlinear Root-finding. 6.2 Rate of Convergence

Chapter 6. Nonlinear Equations. 6.1 The Problem of Nonlinear Root-finding. 6.2 Rate of Convergence Chapter 6 Nonlinear Equations 6. The Problem of Nonlinear Root-finding In this module we consider the problem of using numerical techniques to find the roots of nonlinear equations, f () =. Initially we

More information

APPLICATION OF A CONSTRAINED OPTIMIZATION TECHNIQUE TO THE IMAGING OF HETEROGENEOUS OBJECTS USING DIFFUSION THEORY. A Thesis MATTHEW RYAN STERNAT

APPLICATION OF A CONSTRAINED OPTIMIZATION TECHNIQUE TO THE IMAGING OF HETEROGENEOUS OBJECTS USING DIFFUSION THEORY. A Thesis MATTHEW RYAN STERNAT APPLICATION OF A CONSTRAINED OPTIMIZATION TECHNIQUE TO THE IMAGING OF HETEROGENEOUS OBJECTS USING DIFFUSION THEORY A Thesis b MATTHEW RYAN STERNAT Submitted to the Office of Graduate Studies of Teas A&M

More information

General Vector Spaces

General Vector Spaces CHAPTER 4 General Vector Spaces CHAPTER CONTENTS 4. Real Vector Spaces 83 4. Subspaces 9 4.3 Linear Independence 4.4 Coordinates and Basis 4.5 Dimension 4.6 Change of Basis 9 4.7 Row Space, Column Space,

More information

An Alternative Proof of Channel Polarization for Channels with Arbitrary Input Alphabets

An Alternative Proof of Channel Polarization for Channels with Arbitrary Input Alphabets An Alternative Proof of Channel Polarization for Channels with Arbitrary Input Alphabets Jing Guo University of Cambridge jg582@cam.ac.uk Jossy Sayir University of Cambridge j.sayir@ieee.org Minghai Qin

More information

On the spectral formulation of Granger causality

On the spectral formulation of Granger causality Noname manuscript No. (will be inserted b the editor) On the spectral formulation of Granger causalit the date of receipt and acceptance should be inserted later Abstract Spectral measures of causalit

More information

Chapter 6. Self-Adjusting Data Structures

Chapter 6. Self-Adjusting Data Structures Chapter 6 Self-Adjusting Data Structures Chapter 5 describes a data structure that is able to achieve an epected quer time that is proportional to the entrop of the quer distribution. The most interesting

More information

MMJ1153 COMPUTATIONAL METHOD IN SOLID MECHANICS PRELIMINARIES TO FEM

MMJ1153 COMPUTATIONAL METHOD IN SOLID MECHANICS PRELIMINARIES TO FEM B Course Content: A INTRODUCTION AND OVERVIEW Numerical method and Computer-Aided Engineering; Phsical problems; Mathematical models; Finite element method;. B Elements and nodes, natural coordinates,

More information

Stability Analysis of a Geometrically Imperfect Structure using a Random Field Model

Stability Analysis of a Geometrically Imperfect Structure using a Random Field Model Stabilit Analsis of a Geometricall Imperfect Structure using a Random Field Model JAN VALEŠ, ZDENĚK KALA Department of Structural Mechanics Brno Universit of Technolog, Facult of Civil Engineering Veveří

More information

STATIC LECTURE 4: CONSTRAINED OPTIMIZATION II - KUHN TUCKER THEORY

STATIC LECTURE 4: CONSTRAINED OPTIMIZATION II - KUHN TUCKER THEORY STATIC LECTURE 4: CONSTRAINED OPTIMIZATION II - KUHN TUCKER THEORY UNIVERSITY OF MARYLAND: ECON 600 1. Some Eamples 1 A general problem that arises countless times in economics takes the form: (Verbally):

More information

Comments on Problems. 3.1 This problem offers some practice in deriving utility functions from indifference curve specifications.

Comments on Problems. 3.1 This problem offers some practice in deriving utility functions from indifference curve specifications. CHAPTER 3 PREFERENCES AND UTILITY These problems provide some practice in eamining utilit unctions b looking at indierence curve maps and at a ew unctional orms. The primar ocus is on illustrating the

More information

THE HEATED LAMINAR VERTICAL JET IN A LIQUID WITH POWER-LAW TEMPERATURE DEPENDENCE OF DENSITY. V. A. Sharifulin.

THE HEATED LAMINAR VERTICAL JET IN A LIQUID WITH POWER-LAW TEMPERATURE DEPENDENCE OF DENSITY. V. A. Sharifulin. THE HEATED LAMINAR VERTICAL JET IN A LIQUID WITH POWER-LAW TEMPERATURE DEPENDENCE OF DENSITY 1. Introduction V. A. Sharifulin Perm State Technical Universit, Perm, Russia e-mail: sharifulin@perm.ru Water

More information

UNCORRECTED SAMPLE PAGES. 3Quadratics. Chapter 3. Objectives

UNCORRECTED SAMPLE PAGES. 3Quadratics. Chapter 3. Objectives Chapter 3 3Quadratics Objectives To recognise and sketch the graphs of quadratic polnomials. To find the ke features of the graph of a quadratic polnomial: ais intercepts, turning point and ais of smmetr.

More information

Optimal Power Control in Decentralized Gaussian Multiple Access Channels

Optimal Power Control in Decentralized Gaussian Multiple Access Channels 1 Optimal Power Control in Decentralized Gaussian Multiple Access Channels Kamal Singh Department of Electrical Engineering Indian Institute of Technology Bombay. arxiv:1711.08272v1 [eess.sp] 21 Nov 2017

More information

Mathematics. Polynomials and Quadratics. hsn.uk.net. Higher. Contents. Polynomials and Quadratics 52 HSN22100

Mathematics. Polynomials and Quadratics. hsn.uk.net. Higher. Contents. Polynomials and Quadratics 52 HSN22100 Higher Mathematics UNIT OUTCOME 1 Polnomials and Quadratics Contents Polnomials and Quadratics 5 1 Quadratics 5 The Discriminant 54 Completing the Square 55 4 Sketching Parabolas 57 5 Determining the Equation

More information

UNIT 2 QUADRATIC FUNCTIONS AND MODELING Lesson 2: Interpreting Quadratic Functions Instruction

UNIT 2 QUADRATIC FUNCTIONS AND MODELING Lesson 2: Interpreting Quadratic Functions Instruction Prerequisite Skills This lesson requires the use of the following skills: knowing the standard form of quadratic functions using graphing technolog to model quadratic functions Introduction The tourism

More information

Analytic Geometry in Three Dimensions

Analytic Geometry in Three Dimensions Analtic Geometr in Three Dimensions. The Three-Dimensional Coordinate Sstem. Vectors in Space. The Cross Product of Two Vectors. Lines and Planes in Space The three-dimensional coordinate sstem is used

More information

INF Introduction to classifiction Anne Solberg Based on Chapter 2 ( ) in Duda and Hart: Pattern Classification

INF Introduction to classifiction Anne Solberg Based on Chapter 2 ( ) in Duda and Hart: Pattern Classification INF 4300 151014 Introduction to classifiction Anne Solberg anne@ifiuiono Based on Chapter 1-6 in Duda and Hart: Pattern Classification 151014 INF 4300 1 Introduction to classification One of the most challenging

More information

5 Mutual Information and Channel Capacity

5 Mutual Information and Channel Capacity 5 Mutual Information and Channel Capacity In Section 2, we have seen the use of a quantity called entropy to measure the amount of randomness in a random variable. In this section, we introduce several

More information

I - Information theory basics

I - Information theory basics I - Information theor basics Introduction To communicate, that is, to carr information between two oints, we can emlo analog or digital transmission techniques. In digital communications the message is

More information

Roberto s Notes on Integral Calculus Chapter 3: Basics of differential equations Section 3. Separable ODE s

Roberto s Notes on Integral Calculus Chapter 3: Basics of differential equations Section 3. Separable ODE s Roberto s Notes on Integral Calculus Chapter 3: Basics of differential equations Section 3 Separable ODE s What ou need to know alread: What an ODE is and how to solve an eponential ODE. What ou can learn

More information

The Kuhn-Tucker and Envelope Theorems

The Kuhn-Tucker and Envelope Theorems The Kuhn-Tucker and Envelope Theorems Peter Ireland ECON 77200 - Math for Economists Boston College, Department of Economics Fall 207 The Kuhn-Tucker and envelope theorems can be used to characterize the

More information

Correlation analysis 2: Measures of correlation

Correlation analysis 2: Measures of correlation Correlation analsis 2: Measures of correlation Ran Tibshirani Data Mining: 36-462/36-662 Februar 19 2013 1 Review: correlation Pearson s correlation is a measure of linear association In the population:

More information

Ch 3 Alg 2 Note Sheet.doc 3.1 Graphing Systems of Equations

Ch 3 Alg 2 Note Sheet.doc 3.1 Graphing Systems of Equations Ch 3 Alg Note Sheet.doc 3.1 Graphing Sstems of Equations Sstems of Linear Equations A sstem of equations is a set of two or more equations that use the same variables. If the graph of each equation =.4

More information

Cycle Structure in Automata and the Holonomy Decomposition

Cycle Structure in Automata and the Holonomy Decomposition Ccle Structure in Automata and the Holonom Decomposition Attila Egri-Nag and Chrstopher L. Nehaniv School of Computer Science Universit of Hertfordshire College Lane, Hatfield, Herts AL10 9AB United Kingdom

More information

The Kuhn-Tucker and Envelope Theorems

The Kuhn-Tucker and Envelope Theorems The Kuhn-Tucker and Envelope Theorems Peter Ireland EC720.01 - Math for Economists Boston College, Department of Economics Fall 2010 The Kuhn-Tucker and envelope theorems can be used to characterize the

More information

0.24 adults 2. (c) Prove that, regardless of the possible values of and, the covariance between X and Y is equal to zero. Show all work.

0.24 adults 2. (c) Prove that, regardless of the possible values of and, the covariance between X and Y is equal to zero. Show all work. 1 A socioeconomic stud analzes two discrete random variables in a certain population of households = number of adult residents and = number of child residents It is found that their joint probabilit mass

More information

EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computer Science. CASA-Report April 2016

EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computer Science. CASA-Report April 2016 EINDHOVEN UNIVERSITY OF TECHNOLOGY Department of Mathematics and Computer Science CASA-Report 6-4 April 26 Discrete and continuum links to a nonlinear coupled transport problem of interacting populations

More information

Strain Transformation and Rosette Gage Theory

Strain Transformation and Rosette Gage Theory Strain Transformation and Rosette Gage Theor It is often desired to measure the full state of strain on the surface of a part, that is to measure not onl the two etensional strains, and, but also the shear

More information