Realization of modified cut-set matrix and applications

IEEE TRANSACTIONS ON CIRCUIT r101J. Tow, “Order of complexity of linear active networks,” Proc. IEE (London), vol. 115, pp. 1259-1262! September 1968. r111F. R. Gantmacher, The Theory of Matrzces. New York: Chelsea, 1959. WI P. R. Brvant, “Problems in electrical network, theory,” Ph.D. dissertatibn, Cambridge, England, July 1959. r131S. Seshu and M. B. Reed, Linear Graphs and Electrical Networks. Reading, Mass: Addison-Wesley, 1961. P41 E. J. Purslow, “The formulation of the state variable equations for the analysis and synthesis of linear electric networks,” Ph.D. dissertation, University of London, London, England, September 1967. electronic circuit?,” Proc. 6th Midwest Symp. Circuit Theory (University of Wisconsin, Madison), 1963. [7] D. S. Rane and J. J. Lang, “Formulation of state space eqnations for active RLC networks by digital computers,” 9th Midwest Symp. Circuit Theory (Oklahoma State University, Stillwater), May 1966. foundat,ions of [S] N. Declares and R. Saeks, “Graph-theoretic linear lumped, finite networks,” Co& Elec. Networks (University of Newcastle, Newcastle, England), September 1966. [9] E. J. Purslow and R. Spence, “Order of complexity of active networks,” Proc. IEE (London), vol. 114, pp, 195-198, February 1967. Realization qf Modified P. SUBBARAMI 475 VOL.CT-17,NO.4, NOVEMBER 1970 THEORY, REDDY, Cut-Set Matrix V. G. K. MURTI, SENIOR Abstracf-The concepts of network of departure and padding n-port network are defined and certain important properties of these networks are stated. A necessary and sufficient criterion for the realization of the modified cut-set matrix of a resistive n-port network having a prescribed configuration is given. A new approach to the synthesis of short-circuit conductance matrix of a resistive n-port network with more than (n + 1) nodes is outlined. Necessary and sufficient conditions for the realization of the potential factor matrix of an (n + 2) node resistive n-port network are also obtained. These conditions lead to a simple t&t for the realizability of the modified cut-set matrix of an (n + 2) node n-port network. Examples are worked out to illustrate the application of these results. MEMBER, and Applications IEEE, AND K. THULASIRAMAN is the diagonal matrix of edge conductances of N, then the cut-set admittance matrix Y, of N is given by (1) in which the rows and columns of Y,, correspond to the port edges and those of Y,, correspond to the nonport _ edges. The modified cut-set matrix C of N is defined as [I] cy = Cl - I. INTRODUCTION N THIS paper we investigate the problem of realization of the modified cut-set matrix [l] of a resistive n-port network and its relation to the realization of the short-circuit conductance and’ the potential factor matrices of such networks. We first introduce the notation to be followed. We consider a resistive n-port network N, having no internal vertices. The linear graph G* of N is assumed to be complete and edges with zero conductances permitted. Let T, the port configuration of N, be in T connected parts T,, T,, * - - , T,. The edges of T, called the port edges, are oriented ‘according to the orientations of the ports. Let T be a subgraph of a tree T, of G*. The edges of T, other than those of T are termed nonport edges and may be oriented arbitrarily. Let co= [I2 2 be the fundamental cut-set matrix of G* with respect to T,, where the rows of C, correspond to the port edges and those of Cz correspond to, the nonport edges. If G Manuscript received June 30, 1969; revised April 2, 1970 The authors are with the Department of Electrical Engineering, Indian Institute of Technology, Madras 36, India. Y,zY,;c,. (59 It can be shown that after rearranging the rows and the columns of C, the partition will be as follows. c = [C’ I C”] where d with C’ being the fundamental cut-set matrix with respect to Ti of the complete graph built on the vertices of T,. The columns of Cp correspond to edges joining a pair of vertices in the same’ connected part of T, while the columns of CN correspond to edges joining a pair of vertices in two different connected parts of T. For an n-port .network containing no negative conductances the magnitude of each entry of CN is less than unity. The potential factor matrix [2], [3] K = [i&J of the n-port network N is defined as the n X n matrix, where kii, called the potential factor of port’j with respect to port i, is the potential of the positive reference terminal of port j with respect to the negative reference terminal of port i when port i is excited with a source of unit voltage 416 IEEE and all the other ports are short-circuited. After rearranging its rows and columns, the matrix K can be partitioned as follows. K= K,, Ku Ku . +. K,; -L--K,, KmK,a . . . h-27 -__--~ K,, K,, K,, . . . &r ----- TRANSACTIONS ON THEORY, to T,, and the corresponding matrix to the tree T*, are both equal to NOVEMBER 1970 of N* with respect +n-tI I 1 T Y (4) -+ --- k-l 0 -A- I 0 I I I- - I I 0 I I I : where the rows of Y correspond to the edges of T (port edges) in each case, then the two networks N and N* are identical. where 1) each submatrix Kij, i = 1, 2, . . . , r has entries comprising 1 and 0 only and is uniquely fixed by the configuration of Tj; 2) each entry in a submstrix Kii, j # i is less than unity and greater than zero, except in degenerate cases; 3) all entries in any row of each Kti, j # i are equal. Proof: a) Let Co and CW,refer respectively to the fundamental cut-set matrices of N with respect to T, and T*,, with the first n rows of C,, and C*, corresponding to the edges of T. Then C*, can be expressed as C*, = AC,,, where the transformation matrix A is of the form T n -. I II. NETWORK OF DEPARTURE,~ADDING n-PORT NETWORK, AND REALIZATION OFTHE MODIFIED CUT-SET, MATRIX In this section we first introduce the concepts of the network of departure and the padding n-port network and outline their important proper&s. We then obtain a necessary and sufficient condition for the realization of the modified cut-set matrix of a resistive n-port network having’a prescribed port configuration. A. Network of Departure and Padding n-Port Network We have -1= 10-IIO y 0 =“‘_o C,,GC;, where G is the edge conductance matrix of N. Now, Y; = C:G(C*,)’ = AC,GC(A’ Theorem I: a) Let N represent an n-port network having a port configuration T. Let the cut-set admittance matrix Y, of N with respect to a tree T,, of N, of which T is a subgraph be given by I+--+ yn I YIO Y, = J, -_--/--__ I I 0 I% 0 I I , CIRCUIT ( -where the rows of the n X n matrix Y correspond fo the edges of T, i.e., the port edges. Then the cut-set admittance matrix Y*, of N with respect to any other tree T*, OFwhich T is a subgraph is also equal to Y,. b’> Let N and N* be two (n + k) node realizations, k > 1, of the same n-port shirt-circuit conductance matrix Y with a prescribed port configuration T. Let T, and T*, be two distinct trees such that T is a subgraph ,of both. If the cut-set admittance matrix of N with respect = AY,A’ = Y,. Hence Theorem la). b) It follows from Theorem la) that the cut-set admittance matrix of N* with respect to TO is also equal to y)c0-i 1 YfO where the first n rows correspond to the edges of T. Since the two networks N and N* have the same cut-set admittance matrix with respect to a common tree configuration T,, they must be identical. Definition I-Networlc of Departure: An n-port network with more than (n + 1) nodes is called a network of departure Nd with respect to a real symmetric matrix Y if its cut-set admittance matrix is equal to YIO --I-F 0 lo 1 where the rows of Y correspond to the port edges. We note that the n-port short-circuit conductance matrix of Nd is equal to Y. It follows from Theorem 1 that, given the port configuration and the short-circuit REDDY et d.: REALIZATION OF MODIFIED CUT-SET 477 MATRIX conductance matrix, the corresponding network of departure Nd is unique. Thus for a given n-port network N, there exists a network of departure that is uniquely determined and referred to as the network of departure of the given n-port network N. DeJinition S-Padding n-Port Network: An n-port network is called a padding n-port network N, if its shortcircuit conductance matrix is equal to zero. Ignoring the degenerate case of a network having zero conductance for all its edges, we note that N, should necessarily contain both positive and negative conductantes. We further n+otethat the short-circuit conductance matrix of the parallel combination of N, and Nd, both having identical port configurations and orientations, is equal to that of N,,. We also note that any n-port network N can be considered as the parallel combination of a network of departure Nd and a padding n-port network N, [4]. Since for a given N, NJ is uniquely determined, the corresponding N, is also uniquely fixed. Let the set of vertices in the ith connected part Ti of the port configuration T of N be denoted as i,, i,, . . + , i,,i. Let (gikim), (gitiJp, and (giriJJ refer to the conductances of edges connecting the vertices ik and j, in the network N, the associated padding network N,, and the network of departure Nd, respectively. Now, giLin = (.CTikiJD+ (ClikiJd. (5) &,i (61 Let This property implies that the parallel combination of Nd and N, is always proper, for any N,, satisfying the stated restrictions. c) A necessary and sufficient condition that an n-port network Nd be a network of departure is that (Si*i)d = 0 and k = 1, 2, . . . , ni. (7) Each of the above quantities is thus equal to the sum of the conductances of all the edges joining the kth vertex in T< to the vertices of Ti in the respective network. Then it follows that Siri = (Sil;i)d+ (Siri)~. (9) With this notation we next proceed to list the important properties of Nd and N,. The proofs of all the properties, except the last two, are not given here but may be readily constructed. Properties of Nd: Properties of N,: e) For a given N, the modified cut-set matrix of the corresponding padding network N, is the same as that of N. f) An (n + 1) node-padding n-port network is the degenerate network with all its edge conductances equal to zero. g) If N contains no negative conductances, then in the corresponding padding network N,, (&A 2 0 for all i and j = 1, 2, . . * , r, j # i and k = 1,2, . . . , n,. h) For a padding network N, with (Sibi)D 2 0 for all i and j, j # i and k = 1, 2, . . * , ni, the following relation must be satisfied. I 0 for k, m = 1, 2, * * . , ni, k # m, andi = 1,2, *.. ,r. Proof: Consider N, with all ports except those in Ti short-circuited. Each subtree Ti, j # i is now reduced to a single vertex j. The conductance of a typical edge in this new network is either (gi,i,), or (Siki)n. If we eliminate by generalized star-mesh conversion the edges of the second type, there results a network in which the edges joining vertices ik and i, have a conductance (gitiJl, + (g<bi,,,)cwhere the latter quantity is nonnegative. Now the resulting (nc - 1)-port network constructed on ni vertices has a short-circuit conductance matrix identically equal to zero. Hence by virtue of Propertyf), we have Since (gib<,,,)cis nonnegative, we have from the above, bd, 5 0. i) CSiri)P a) The modified cut-set matrix of Nd does not exist. b) The n-port short-circuit conductance matrix of the parallel combination’ of N, with any other n-port network N, having a determinate short-circuit conductance matrix and the same port configuration and orientation as N,, is equal to the sum of the short-circuit conductance matrices of NJ and N,. (10) d) If an n-port network N contains no negative conductances, then in the corresponding network of departure Nd, the conductance of every edge joining the vertices in the same connected part of T is nonnegative. (Oids = % giLin, for all i, j = 1, 2, . . f , r, j Z i, Given a padding 2 0 foralliand n-port network N, such that j = 1, 2, .a. r, j # i, and k = 1,2, . . . , ni then a network of departure Nd can always be obtained such that the parallel combination of N, and N, contains no negative conductances. Pioof: First we note that for the given padding n-port network N,, 478 IEEE foralli (SiriJt, 5 0 = 1,2, ... ,r, and k, m = 1,2, . . . , ni, k # m by’virtue of Property h). We proceed to prove Property i) by giving a procedure for constructing a suitable Nd. Construct a network N, having the same port configuration and orientations as N, and such that (gw,L for all k, m = 1, 2, . . . , n,, = - (sd, k#m,andi=l,2,...,r forall i,j and (11) = 1,2, a.. ,r, j f i, k = 1,2, .a. ,ni, m = 1,2, .** ,nj G3 where For the network N, constructed as above, TRANSACTIONS ON CIRCUIT THEORY, NOVEMBER 1970 combination of the given N, and the network of departure N, constructed to satisfy (11) and (12) contain no negative conductances. B. Realization of the Modi$t$ Cut-Set Matrix of an n-Port Network Let there be given a real matrix C partitionable as in (3) and such that C”, i =*l, 2, * * . , r is realizable as the fundamental cut-set matrix with respect to Ti, i = 1, 2, . . . , r. Consider an n-port network N having a port configuration definedby Ti, i = 1,2, *** , r. The graph of N is assumed to be complete. Each edge of N corresponds to a unique column of C and this correspondence is assumed to be known. It is required to obtain a necessary and sullicient condition that C represent the modified cut-set matrix of an N containing no negative conductances. In the light of Properties e)-i), the above problem can be. considered equivalent to obtaining necessary and sufficient conditions for the realization of C by a padding n-port network N, whose conductances satisfy the property stated in Property i). Thus, we have the following theorem on the realization of the modified cut-set matrix C. Theorem 2: If and only if there exists a real diagonal matrix G, such that CG,C: = 0 CG,,C; = 0 det [C,G,C:] # 0 ni = qq (SixiJl, + (Siti)z3 2 O (fLri)s and for all i, j = 1, 2, ... and = 0, foralli and j = 1, 2, ... ,r, . and k = 1,2, .‘. ,ni. j # i, (13) It’ follows from Property c) and (13) that the network N, constructed with its conductances satisfying (11) and (12) is a network of departure N,. The conductance girim in the parallel combination of N, and N, is given by I ,r, j # i, k = 1,2, ..a ,ni, the matrix C can be realized as the modified cut-set matrix of an n-port network containing no negative conductances. Proof-Necessity: This follows from Properties e)-i) and Theorem 4 of [3]. Proof-Su&iency: This follows from Properties e)-i) and Theorems 4 and 6 of [3]. The usefulness of the results of this section in the synthesis of the short-circuit conductance matrices of resistive n-port networks will be considered in the next section. III. SYNTHESIS OF THE SHORT-CIRCUIT CONDUCTANCE MATRIX OF A RESISTIVE n-PORT NETWORK for all i, j = 1, 2, *.f ,r, k = 1,2, ... j#i ,ni and, m = 1,2, *** ,nj. (14). Also Q&in. = (SC*iA + (giki,>d~ = 0 foralli; 1,2, . ..‘.T, k,m=J,2,,-;.,ni, k#m-. (li) From the above equations it follows cthat the parallel* The problem of realization of the short-circuit conductance matrix Y of a resistive n-port network has received the attention of research workers in network theory for more than a decade. Whereas the problem of synthesis of n-ports with (n + 1) nodes is considered solved, the problem of synthesis with more than (n + 1) nodes is yet unsolved. Extending Cederbaum’s approach for (n + 1) node network-synthesis, one approach to the above problem may be to decompose the given matrix’ Y as Y = CGC’ where C is a real matrix and G is a real diagonal .matrix of nonnegative numbers, and then to realize’ C as the modified cut-set matrix of a resistive n-port network. The successful. application of this pro- REUDY et al.: REALIZATION OF MODIFIED CUT-SET 479 MATRIX cedure, however, requires as a first step, a procedure for the decomposition of Y, which is not known. Further, it is known that such a decomposition is not unique. A slightly less difficult problem is to consider the realization of a real symmetric matrix as the short-circuit conductance matrix of an n-port network having a specified port configuration. The earliest approach suggested by Guillemin [4] to solve this problem essentially requires the determination of a) the unique network of departure Nd with respect to Y having the prescribed port configuration, and b) a suitable padding n-port network N, so that the parallel combination of N, and Nd contains no negative conductances., The procedure given by Guillemin to generate padding n-port networks is general, in that it can generate all possible padding n-port networks. However, in the light of the discussions of the last section, it follows that only the class of padding n-port networks whose conductances satisfy the property stated in Property i) are required. Hence, while applying Guillemin’s procedure this constraint on conductances should be incorporated. Other approaches due to Frisch and Swaminathan [5] and Halkias and Lupo [6] can also be stated in terms of networks of departure and padding networks. They . differ from that of Guillemin only in the method used to generate padding networks. In generating a padding network, Frisch and Swaminathan first obtain the cut-set admittance of the padding network in terms of n(r - 1) + [r(r - 1)]/2 ar b i t rarily assumed numbers and then obtain the edge conductance values from the cut-set admittance matrix. A significant result of this investigation is the derivation of the supremacy condition that is necessary for the realizability of the Y matrix of (n + 2) noderesistive n-port networks. This condition is simple and easy to apply. Halikias and Lupo have obtained formulas that express directly the edge conductances of (n + 2) node-padding n-port networks in terms of (n + 2) arbitrarily assumed numbers. The extension of this approach to generate padding networks with more than (n + 2) nodes, given later by Lupo [7], is, however, not general, since it generates only a class of padding networks whose potential factors are related in a special manner. In this section we present an alternate procedure for Y-matrix synthesis, which again, differs from that of Guillemin only in the method used to generate padding networks. The procedure is as follows. Step 1: For the given Y matrix obtain the unique network of departure Nd having the prescribed port configuration. Step 2: Assuming suitable values for nonzero nonunity potential factors and using Theorem 2 of [2], construct the modified cut-set matrix C appropriate to the given port configuration. Obtain a padding n-port network N, such that (siri); 2 0 for all i and j,’ j # i, i, j = 1; 2, . *,* r and k = 1, 2, . . * , ni. For this the procedure ,contained in Theorem 2 may be used. Step 3: If, for one assumed C, no N, can be obtained such that the parallel combination N of N, and Nd con- tains no negative conductances, assume a different set of values for the nonzero nonunity potential factors and hence a different modified.cut-set matrix. Repeat Step 2. Step 4: If an N, can be found, which results in an N containing no negative conductances, then this network N will be a proper realization of Y. If not, the matrix Y isnot realizable by resistive n-port networks having the prescribed port configuration. A significant feature of the method used in Step 2 of the above procedure to generate padding networks is that edge conductances of padding networks are expressed in terms of potential factors. This will be helpful in extending the results of resistive n-port synthesis to the synthesis of Y matrices of, RLC networks, since, in the synthesis of the latter, all residue matrices, if they are real, or all parameter matrices are required to be realized by networks having identical potential factors [8]. This new method to generate padding networks requires the assumption of suitable values for nonzero nonunity potential factors. Since these potential factors should be the same as those for the final network N, to avoid excessive computational labor, it is necessary that we obtain necessary and sufficient conditions, which ‘the nonzero nonunity potential factors of resistive n-port networks containing no negative conductances should satisfy.. This leads us to the problem of synthesis of the potential factor matrix of resistive n-port networks. In the next. section we consider this aspect of the Y-matrix synthesis problem in relation to an (n + 2) node network. IV. REALIZATION OFTHE POTENTIAL FACTORMATRIX OF AN (n f 2) NODE-RESISTIVE n-Ponr NETWORK In this section we obtain a necessary and sufficient condition which the elements of K,, and K,, should satisfy in order that h, KzA I ’ : be. realizable as the potential factor matrix of an (n + 2) node-resistive n-port network whose port configuration defines Kll and K,,. It can be shown that such a condition provides an effective criterion for the realization of the modified cut-set matrix of (n + 2) node-resistive n-port networks. We also illustrate the application of this condition,to-the synthesis of the short-circuit conductance matrix of (n + 2) node n-port networks. We first consider a resistive (n + 2) node n-port network N containing no negative conductances. We assume, without loss of generality as will be shown later, that T, and T, of N are linear trees. In designating the vertices, edges, etc., of N, we adopt a notation different from the one used earlier and simpler in the ‘present context. Let the vertices of any linear tree To of ,.N, of -which T, and T, are subgraphs, be numbered consecutively starting from one end vertex of To. Let there be m edges in T, and (n - v) edges in T,. Let eii represent the edge-connecting vertices i and j. Let ei,i+l denote port i if i _< m and ei+l,,+z denote port i if i 2 m -I- 1. em+l,m+2will then 480 IEEE Fig. 1. Notation followed in vertex designation, represent the nonport edge of To as shown in Fig. 1. Let the orientation of eii ,be away from j, if j > i, and let gii refer to the conductance of edge eii. Let TRANSACTIONS ON CIRCUIT i I i.2 = j=ji12 gijt si.l WL+1 = C gii7 j==l lc,, = ki, iim or i>m+2. and j,k n+* W&+1 s =. c a.2 = c xi.,. i=m+2 i=l i 2 m + 1 and 07) 2 m+ 1 j, lc 5 m. (21) Let ki = lCis,+, = kc,,+2 = ” * = lCi,n and lc< = ki.1 = lc,,Z = . * * = lci., Let 1970 its rows arranged in the same order as the columns of co.:, cb,i, etc. We have, as already mentioned in Section I, that WZ+ 1 (16) NOVEMBER port and edge orientation. 7L+2 s THEORY, ilm (22) ikm+l. It follows from Theorem 4 of [3] that the diagonal matrix G of edge conductances of N satisfies Let the modified cut-set matrix C and the submatrix C, of the fundamental cut-set matrix C, be partitioned as CGCI = 0 i.e., where the rows of C. and C,, correspond to the port edges in T1 and those of C, and Clb correspond to the port edges in T2. Let the columns of C, be rearranged and partitioned as where C,,< consists of all columns of C, corresponding to the edges joining vertex i in T, to all vertices in TP; C,. Ti consists of all columns of C, corresponding.to the edges connecting two vertices in Tie Similarly the columns of C,, Ctb, C,,, and C:! can be arranged and partitioned. We note that C b.T, -- C 1o.T. = 0 ca,T, 0 C1b.Tx C 2.T. = 0, = = 0 (20) c2.T. = 0 and that all the entries in each of the row matrices C,,i, i = 1,2, e-0 ) m+ lareequaltol. Let gi be the column matrix of conductances of edges corresponding to the columns of C,, 0 C,. <, etc., with (23a) C,GC; = 0. (23b) and It can be shown from (23a) that m+l c si.2 i= ki = j=Ys Thus, for a network have from (24) 1,2, ... of nonnegative ,m. (24) conductances, k, > k, > lc, 2 * . ’ 2 lc,. we (25) Also it can be shown from (23b) that *+2 Ic, = 0 C,GC; = 0 C i-i’> xi,1 i=m+l,m+2;.. ,n (26) and it follows that ii m+l 2 hn+2 2 ka,, 2 . - ’ 2 k. (27) Hence the elements of the submatrices K,, and K,, of the potential factor matrix K of an (n + 2) node-resistive n-port network having a linear 2-tree port configuration and no negative conductances must satisfy (21), (%), REDDY et cd.: REALIZATION OF MODIFIED CUT-SET 481 MATRIX CGC: = 0 and (27). We note that K,, and K,, for such a network will have the following form. 1 1 1 **. 1 1 Kz2= 1 1 .*. 1 1 ..* ... Pf3) Next we show that if the elements of the submatrices K,, and KP, satisfy (21), (25), and (27) and if K,, and \ K,, are of the forms specified in (28), then represents the potential factor matrix of an (n + 2) noderesistive n-port network containing no negative conductances and having a linear 2-tree port configuration T defined by I<,, and K,,. Using K, we first construct the modified cut-set matrix C appropriate to the port configuration T. This can always be done in view of Theorem 2 of [2]. We then show that it is possible to construct a padding n-port network N having the port configuration T and the modified cut-set matrix C and satisfying the constraint si.2 2 0 for all i I si.1 2 0 for all i 2 m + 2. i 5 m + 1 (294 Xi.1 2 0 for all i 2 m + 2. (294 (304 c,Gc; (3W = 0 CaGC:. = 0 (3Oc) c,Gc:b = 0 (304 C,GC:, = 0 (304 si.2 2 0 for all ilmfl Wf) xi,, 2 0 for all i2m+2 Wd xi.2 > 0 forsome i _< m + 1. @Oh) From (24), which can be obtained from (30a), we get the following Xi,2 = (k;-1 - ki)S, i = 2,3, . -. ) m s,,, = (1 - ICJS (31) ‘.‘, S m+.,,2= k,S. Next consider (30b) and (30e). After ,setting Cb, T, = 0, C2.Tl = 0, and Cz.Tn = 0, (30b) can be written as Cb.l{gl} + *’ ’ + Cb,i{gi) f We obtain from (294 ei.m+4 1 - JGm+, 1 - k,,, “’ +c CGC: = 0 -km+, for all C,GC: = 0 C*,*b71} + **’ ei,m+3 si.Z 2 0 m + I For this we adopt the procedure contained in Theorem 2. Let G be the diagonal matrix of edge conductances of the required padding n-port network. We seek a solution for G of the following equations. ei.m+z (29 4 We note that det [C2GC;] # 0 if (29d) and (29e) are satisfied and if further Xi,, > 0 for some i i m + 1. We also note that if C,GC:, = 0 then CbGC:, = 0 since CCC{ is symmetric. Thus, (29a-e) are equivalent to the following. 1 1 1 1 n-mrows. .................. .................. 0 l-0 0 0 ..* 0 1 1 o o 1 det [CzGC:] f 0 1 ... 1 1 0 1’1 ..I 1 11 mrows K,, = o o 1 . . . . . . . . . . . . . ..I.. .................. 0 1 -0 0 0 ..’ GW f (30e) the following Cb,i{gi} = 0 = b.m+l~gm+lf i= 0. (32) sets of equations 1,2, v.-,m. (33) Simultaneous solution of (32) and (33) yields the following. Cb,i{gi) = 0 i = 1, 2, . . . , m + 1. (34) We note that Cb,l = Cb,, = ** * = Cb,,+,. After expressing Cb,i in terms of potential factors, (34) becomes ..- ei,n+2, . *. 1 - k,,, 1 - k,,, . * * 1 - I%,+, - kn+2 -JL+, ........................................ ........................................ -k, ... 1 : k, lb -k, -k, gi.m+2 Qi,m+3 gi,n+z = 1 0 i = 1,2 7 -f- ,m+ 1. (35) 482 IEEE We obtain from the above L k, 1,2, .a. ,m+ 1. (36) i = 1,2, ..a ,m+ - kj.4)&,2 I and gi.m+z = (1 - k,+,)Si,z i = 1, 2, ... ,m + 1 gi .n+2 = kSi.z i = 1,2, ... , m + 1. (37) Consider, (31) and (37). Choosing any positive value for S, the values obtained for Si.2 from (31) will be nonnegative, since k, 2 k, > . . . > k,. Further, at least two of them will be positive. Hence, such a choice of Si,, will satisfy (30f), (3Og), and (30h). We then use these values for AS’;,,and (37) to calculate the values of conductances of edges connecting vertices in T, to vertices in T,. We note that the conductances of these edges will be nonnegative since 1 2 k,+l 2 k,+a 2. . * 2 k,. The following expression for conductances of edges connecting vertices in T1 to vertices in T, can also be obtained using (26) and (37). Si.2&,1 s for all and i in T, (33) jinTz. It only remains to obtain the conductances of edges connecting the vertices in T1 only and also of the edges connecting vertices in T, only. gii = i = 1,2, ... ,m+ j=l,2;..,m+l, - ki-,)k”S i = m+3, - k<-,)S, i, j = m + 3, m + 4, .a. ,n + 1, j # i. -S m+2,1Si.l Qm+z.i = S = ~(1 - k,+l)(k+-2 - kj-,>S m+3;-- ,n+ 1 (3% a.2 2 0 for all i 5 m + 1 a.1 2 0 for all i 2 m + 2 xi.2 > 0 forsome and Step 1: Choose any using (31) all positive i_< m+ value for 1. S. Obtain i = I,2 7 . . . , m + 1. Xi,2 Step ,%?:Obtain the conductances of the edges connecting vertices in T, to those in T, using (37). Step S: The conductances of edges connecting any two vertices in T, and of edges connecting any two vertices in T, can be obtained using (39). It follows from the above and Theorem 2 that an (n + 2) node-resistive n-port network containing no negative conductances, having a linear 2-tree port configuration defined by K,, and Kz2, and having the modified cut-set matrix C can always be constructed. Since matrix C was obtained from K = Ku K,2 i K21 Km K = K,, LKz, i= ... I it follows from Theorem 9 of [3] that such a network will have K as its potential factor matrix, when K,, and K,, satisfy (21), (25), and (27). Thus, we have the following theorem on the realization of the potential factor matrix of (n + 2) node-resistive n-port networks. Theorem 8: Let a real matrix j#i S - ki-l)(ki-2 S 1 -&.,&,I = -(kid2 l%‘o Summarizing the results of the above discussions, we outline below the steps to be followed in obtaining a padding n-port network iV having a linear 2-tree port configuration and the modiied cut-set matrix C and satisfying the constraint j=m+3;-.,n+l S NOVEMBER = -(1 - k,+AS. Solving (36) we get, -s;,,si~, THEORY, gm+2.*+2 = -&n+2.1&+2.1 S i: gii = CIRCUIT and 1C m*2 e2 . i= = = -(ki-, k ??I+1 gii = ON -&.lSn+2,1 gi.n+2 gij = (kid TR4NSACTIONS ,n+l K,s 1 K22i -be given. It,can be realized as the potential factor matrix of an ,(n +’ 2) node-resistive ,n-port network containing no negative conductances and having a linear 2-tree REDDY et al.: REALIZATION 0~ 48.3 CUT-SETMATRIX MODIFIED as the modified cut-set matrix of a resistive 4-port’ network having the port configuration shown in Fig. 2. If C* is realizable by a 4-port network N* having the required port configuration, then the modified cut-set matrix C of the 4-port network N constructed on N* and having the linear 2-tree port configuration shown in Fig. 3 can be obtained as C = A’C* where the matrix port configuration defined by K,, and K,, if and only if the entries of the submatrices K,, and KS, of K satisfy the foll.owing conditions. kii = k,, = ki i<m and j, k = m + 1, m + 2, ... ,n, 1 0 1 0 0 or i 2 m + 1 and j, k = 1, 2, ..a , m, and relates the port voltage vector V*, of N” to the port voltage 1 > k,,, The potential factor matrix Theorem 2 of [2] is given by 2 k,,., 2. km+3 L . . . 2 k,. Consider an n-port network N having a port configuration T that consists of two parts T, and T,. Construct on the network N another n-port network N* such that the port configuration T* of N* consists of two parts TT and T*, with T$, i = 1, 2 constructed on the same vertices as Tie Then the potential factor matrix K* of N* can be obtained from the potential factor matrix K of N without making use of the conductance values of N. That is, the realizability of K* can be tested by testing the realizability of K and vice versa. The process of transformation from K* to K is not difficult. In view of these we can conclude that the necessary and sufficient conditions stated in Theorem 3 are general ones applicable to testing the realizability of the potential factor matrix of any (n + 2) node-resistive n-port network. We next illustrate the application of the above results in the realization of the modified cut-set matrix and Y matrix of (n + 2) node-resistive n-port networks. Example 1: Let it be required to realize the matrix ‘1 c* = 1 1 0 (j 01010-10 00101 .o 0 0 0 0 0 0 From K, we get K, = 5, K, = g, K, = 8, and K, = 9. It may be verified that the entries of K,, and K,, satisfy the conditions stated in Theorem 3. We next obtain a padding 4-port network having the above potential factor matrix K in which all Sr.a are greater than zero. Choose S = 9 and obtain S1,2, A!?&,~,S3.2, and Sd,, using (31) as S I.2 --2 1 -Q s2.2 = 2 s3.2 = 2 s,,, = 3. Then using (37) and (39) the edge conductance matrix G, of the required padding 4-port network N, is obtained as 3 -6 fC Q+ 10 0 $ K of N obta ,irled, using -$ $ Q Q -+ -$ -6 g f -f $ -# -4 Q 4 -8 -Q 3 -Q t -Q Q -8 Q1 484 IEEE G, = diag ( g12 g13 g14 g15 -Q -Q # =diag{-+ TRANSACTIONS gle q23 gz4 g25 ga6 g34 q -6 -$ # Lg -Q gas i ON CIRCUIT THEORY, g36 ge5 g4e q .Lg Lg NOVEMBER 1970 gs6) -zo 91. Next a suitable network of departure Nd is obtained, using the procedure stated in the proof of-Property i), as Gd = diag I q12 g13 g14 g15 g16 g23 g24 g25 ga6 g34 ga5 g36 g45 g46 g561 = diag { Q Q 8 0 0 i$ $ 0 0 G The 4-port network N obtained as the parallel combination of N, and Na and having the linear 2-tree port configuration shown in Fig. 3 will have C as its modified cut-set matrix and hence the 4-port network N* constructed on 21Jand having the port configuration shown in Fig. 2 will have the given C* as its modified cut-set matrix. The 4-port network N* is shown in Fig. 4. Example W: Let it be required to realize the matrix y = 61 45 16 5 45 95 40 16 40 56 5 1 0 0 351 : 5 5 0 0 -s2p2 923 9 -82 16 2 5 2 -5 916 = K,&,, 925 = (1 - &)X2,2 9 26 = K4S2s2 >_ 935 = (1 g36 = K,S3,2 945 = (1 cl 46 = K,S,., 912 = g13 = - 2 0 2 -5 0 K,)s3,2 -26 > -24 9}# 2s4.2 - -&.2&.2 > -l6 = s - where s = (81.3 + s2s2 + x3.2 + s4.2). A proper choice of K,, Xl,,, S,.,, S,,,, and S,,, should be made so that the above constraints are satisfied. Further, the following conditions should be satisfied: -5 5 26 Next we have to generate a 4-port padding network N, having the specified port configuration and such that the parallel combination of N, and N, contains no negative conductances. Thus we obtain the following constraints on the edge conductances of N,. 915 = (1 - kJ&,2 2 s all Si., should be greater than zero, 1 > K, 2 K, 2 K, K4 < 1. Two such choices are as follows: Gd = diag (Q12 Q13 914 6’15 Q16 923 Qm Q2s Q26 gsa Q3s 29 zo 0 = 24=-- 9 34 as the short-circuit conductance matrix of a 4-port resistive network having a linear 2-tree port configuration shown in Pig. 5. The edge conductance matrix G, of the network of departure with respect to Y and having the required port configuration is obtained as = diag (16 0 24 0 0 16 5 g36 -5 Q45 0 g46 gsej 0 35}. K, = 5/10 S 1.2 = 30 s,,, = 30 s,,, = 30 s,,, = 20 x23 = 10 x,.2 = 20 x4.3 and K, = 5/10 S 1.2 = 20 = 10. Using the above values suitable padding 4-port networks can be obtained. The final 4-port network corresponding to the first choice and realizing the given Y matrix is shown in Fig. 6. V. CONCLUSIONS > - kJS4.2 2. 0 2 0 -&,2s3,2 > s - s - -s,.,s3,2 > g5,, = -(l - -‘1,2’4.2 g14 5 -29 2 -35 K,)K,S > = s -16 - -16 In this paper we have considered the realization of the modified cut-set matrix of resistive n-port networks and certain aspects of its relation to the Y-matrix synthesis problem. Based on the properties of network of departure and padding n-port network, certain necessary and sufficient conditions are given for the realization of the modified cut-set matrix of n-port networks having a prescribed port configuration. The new procedure suggested for the realization of the short-circuit conductance matrices of n-port networks having more than (n + 1) nodes diiers from that of REDDY et cd.: REALIZATION OF MODIFIED Port configuration Fig. 2. Fig. 3. CUT-SET of network 485 MATRIX required in Example Linear a-tree port configuration used in Example 1. Fig. 5. Port configuration Network realizing the modified Example I. cut-set matrix C* of Guillemin [4] in the method adopted to generate padding networks. The method used in this paper expresses edge conductances of padding networks in terms of potential factors. A first step towards systematizing the Y-matrix synthesis procedure suggested in this paper is to obtain the necessary and sufficient conditions that the potential factors of n-port network containing no negative conductances should satisfy. Such a condition has been obtained in this paper, for the special case of (n + 2) node networks. Though a similar condition can be shown to be necessary in the general case, its sufficiency is yet to be established. In the meantime, the application of the results obtained to the synthesis of the Y matrix of an (n + 2) node n-port network has been illustrated. We wish to point out that the formulas expressing the edge conductances of (n + 2) node-padding n-port networks, as obtained in this paper, can be given in terms of all potential factors and S, or all Si,2 and JS,,~,or all SiVa and all K, i=m+1,m+2,-**,n all S,,1 and all K, i = 1,2, . . . , m. required in Example 2. 1. Fig. 6. Fig. 4. of network or Such expressions will be useful, as stated in Section III, to extend these results to the synthesis of RLC n-port networks. We note that the potential factors SieZ and S,,l and S are all the same both for the padding network and the final network that will be the parallel combination of the padding network and a suitable network of departure; Thus, the arbitrary quantities that are assumed in gen- Network realizing the Y matrix of Example 2. erating a padding network can be identified with the corresponding quantities in the final network required. The procedures used in [5] and [6] do not permit such straightforward identifications. Another result of importance that requires special mention is that if a network is to be the padding network of an (n + 2) node-resistive n-port network containing no negative conductances, then the conductances of the edges in this network connecting vertices in T, to vertices in T, should be nonnegative. This follows from the nonnegativeness of Si.‘L and S,,l. All the procedures presented so far (including the present one) for the synthesis of the Y matrix of resistive n-port networks [5], [6] essentially aim at generation of suitable padding networks. Therefore, it is quite possible to start from one set of formulas for edge conductances of padding networks and obtain another set. Even though some of the procedures given in this paper require further study and systematization, it is hoped that the theory developed in this paper provides more insight into several aspects of analysis and synthesis of n-port networks. The authors have obtained certain necessary conditions and certain sufficient conditions for the synthesis of the Y-matrix (n + 2) node-resistive n-port networks. Further, the application of the techniques discussed here has led to some results on the lower bound on the number of conductances and amount of conductance required for the realization of a real symmetric matrix as the short-circuit conductance matrix of n-port networks having more than (n + 1) nodes. REFERENCES [l] I. Cederbaum, “On equivalence of resistive n-port net.works,” IEEE Trans. Circuil Theory, vol. CT-12, pp. 3:18-34-l, Septemher lC~f?.i. .._. -. __. [2] V. G. K. Murti and K. Thulasiraman, “Parallel connection of n-port networks,” F’roc. IEEE (Letters), vol. 55, pp. 13161217, July 1067. [3] K. Thulasiraman and V. G. K. Murti, “Modified cut-set matrix 486 IEEE TRANSACTIONS Proc. IEE (London), vol. 115, pp. of an n-port network,” 1263-1268, September 1968. [4] E. A. Guillemin, “On the realization of an nth order G matrix,” IRE Trav. Circuit Theory, vol. CT-S, pp. 318-323, September 1961. [5] K. R. Swaminat,han and I. T. Frisch, “Necessary conditions for the realizability of n-port resistive networks with more than Trans. Circuit Theory, vol. CT-12, (n + 1) nodes,” IEEE pp. 520-527, December 1965. Connectivity ON CIRCUIT THEORY, NO. 4, NOVEMBER 1970 [6] F. J. Lupo and C. C. Halkias, “Synthesis of n-port networks on two-tree port-structures,” IEEE Trans. Circuit Theory, vol. CT-12, pp. 571-577, December 1965. [7] F. J. Lnpo, “The synthesis of transformerless n-port networks on mnltitree port st,ructnres,” IEEE Trans. Circuit Theory, vol. CT-15, pp. 21 l-220, September 1968. [8] K. Thulasiraman and V. G. K. Murti, “Synthesis applications of modified cut-set matrix,” Proc. IEE (London), vol. 115, no. 9, pp. 1269-1274, September 1968. Considerations in Survivable Networks H.FRANK, VOL. CT-17, Design MEMBER,IEEE,ANDW.CI~OU,MEMBER;IEEE Abstract-The problem of constructing networks that are “survivable” with respect to branch damage is considered. The networks are modeled by linear graphs and a square symmetric “redundancy” matrix R’ = [rii] is specified. AIgorithms are given to construct au undirected graph G with a minimum number of branches such that 1) G contains no parallel branches, and 2) for all i, i there are at least r$i branch disjoint paths between the ith and $h vertices. These algorithms are complicated but may easily be applied to construct graphs with several hundred yertices. INTRODUCTION consideration in the design of many ,PRIME networks is that the network “survive” an enemy ,A attack. The desired type of survivability depends on the nature of the‘threat and the function of the network. In a number of cases, a reasonable measure of survivability is the number of links and/or stations that must be desstroyed before communication is disrupted. Networks may be modeled by linear graphs so that stations correspond to vertices and links correspond to branches. With this model, a number of authors have studied various aspects of analysis and synthesis with survivability as a parameter [l]-[lo]. Some of the synthesis studies have been concerned with maximizing the minimum number of vertices and/or branches that must be removed from the graph before the graph becomes disconnected [l]-[5]. Steiglitz et al. [6] have considered the general .minimum-cost synthesis problem and have obtained many important results using heuristic programming. Other studies have concentrated on developing efficient methods of finding the minimum numbers of vertices and/or Manuscript received July 25, 1968; revised March 10, 1970. This research was supported by the Execntive Ofice of the President, Office of Emergency Preparedness, Washmgton, D. C., &hile H. Frank was on leave from the University of California, Berkeley, and by the Network Analysis Corporation, Glen Cove, N. Y. The authors were with the Department of Electrical Engineering and Computer Sciences, University of. California, Berkeley. They are now with Network Analysis Corporation, Glen C@ve, N. Y. 11542. branches that must be removed to separate specified pairs of vertices. In this paper, the problem of constructing networks survivable with respect to branch damage is examined. Such damage could be of particular concern in networks subject to sabotage or other threats to which stations are relatively invulnerable. Specifically, suppose an n’ X n’ symmetric matrix R’ = [rii] of nonnegative integers is specified such that rii = 0 for i = 1, * . * , n’. The problem considered is: construct ati undirected graph G without parallel branches and with fewest branches so that there are at least r:j branch disjoint paths between the ith and jth vertices for i, j = 1, 2, . . . , n’. In the communication network corresponding must be destroyed ith and jth stations to such a graph, at least r: i branches before communication is totally between the disrupted. PRELIMINARY CONSIDERATIONS Let G be an undirected graph without parallel branches, defined by the ordered pair (V, I’) (written G = (V, I’)). cv = {vl, * * * ) v,) is the set of vertices of G, and I? is the set of branches defined by I’ = ([vi, vi]} such that [v,, v,] E r if and only if there is a branch between v, and vL. Branch [v,, v,] E r is said to bi: incident at vertices v. and vl, and v, and vL are said to be adjacent. Associated with [vi, vi] is a capacity c[vi, vi] representing the total amount of allowable flow in [vi, vi]. The maximum possible flow between vi and Vj is known as the terminal capacity tij and can be determined by the max-flow mincut theorem [ll]. Then X n matrix T = [ti,] where tij = 0 fori = 1, -.- , n is called the terminal capacity matrix [12]. Since G is undirected, T is symmetric. The degree d(i) of a vertex vi is the number of branches incident at that vertex. Given G, it is easy to find vertex degrees. A less trivial problem is: given a set of positive integers k,, I&, . . . , k,, construct an n-vertex graph G .wit,hout parallel branches whose vertices have degrees j&. . .:, k,. Not all sets of positive integers can be realized