327 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-21, NO. 3, MAY 1974 REFERENCES [l] M. A. Maidique, “A high-precision monolithic super-beta operational amplifier,” IEEE J. Solid-State Circuits, vol. SC-7, pp. 480-487, Dec. 1972. [2] P. R. Gray, “A 15-W monolithic power operational amplifier,” IEEE J. Solid-State Circuits, vol. SC-7, pp. 474480, Dec. 1972. [3] G. Meyer-Brotz and A. Klev. “The common-mode reiection of transistor differential amplil%rs,” IEEE Trans. Cirwa’t Theory, vol. CT-13, pp. 171-175, June 1966. [41 G. Erdi, “Common-mode rejection of monolithic operational amplifiers,” IEEE J. Solid-State Circuits (Corresp.), vol. SC-5, pp. 365-367, Dec. 1970. 151 R. J. Widlar, “Design techniques for monolithic operational amplifiers,” IEEE J. Solid-State Circuits, vol. SC-4, no. -^ 184-191. Aug. 1969. [61 T. M. Frederiksen, W. F. Davis, and D. W. Zobel “A new current-differencing single-supply operational amplifier,” IEEE J. Solid-State Circuits, vol. SC-6, pp. 340-347, Dec. 1971. [71 HA-2600/2602/2605 Operational Amplifier Data Sheet, Harris Semiconductor Corp., June 1971. [81 J. E. Solomon, W. R. Davis, and P. L. Lee, “A self compensated monolithic operational amplifier with low mput current and high slew rate,” ISSCC Digest Tech. Papers, pp. 14-15, 1969. 191R. W. Russell and J. E. Solomon: “A high-voltage monolithic operational amplifier,” IEEE J. Solid-State Circuits, vol. SC-6, pp. 352-357, Dec. 1971. HOI “A low drift, low noise monolithic operational amplifier for low level signal processing,” Fairchild Semiconductor, Appl. Brief 136, July 1969. 1111W. E. Hearn, “Fast slewing monolithic operational amplifier,” IEEE J. Solid-State Circuits, vol. SC-6, pp. 20-24, Feb. 1971. [12] H. C. Lin. “Comoarison of invut offset voltage of differential amplifiers using bipolar transistors and field-elect transistors,” IEEE J. Solid-State Circuits (Corresp.), vol. SC-5 pp. 126-128, June 1970. [1311 B. A. Wooley, S. J. Wong, and D. 0. Pederson, “A computeraided evaluation of the 741 amplifier,” IEEE J. Solid-State Circuits, vol. SG6, pp. 357-366, Dec. 1971. [I41 T. J. van Kessel, “An integrated operational amplifier with a ~;ov;l HF behavtour,” ISSCC Digest Tech. Papers, pp. 22-23, [15] H. Krabbe, “Stable monolithic inverting op amp with 130 V/ps slew rate,” ZSSCC Digest Tech. Papers, pp. 172-173, 1972. [16] R. C. Dobkin, “LM 118 op amp slews 70 V/j~s,” National Semiconductor Linear Brief 17, Aug. 1971. [17] F. D. Waldhauer, “Analog integrated circuits of large bandwidth,” in 1963 IEEE Conv. Rec., Part 2, pp. 200-207. [18] R. J. Apfel and P. R. Gray, “A monolithic fast-settling feedforward operational amplifier using doublet compression techniques,” ISSCC Digest Tech. Papers, pp. 134-155, 1974. [I91 P. C. Davis and V. R. Saari,. “High slew rate monolithic operational amplifier using compattble complementary PNP’s,” ZSSCC Digest Tech. Papers, pp. 132-133, 1974. Maximum Power Transfer in n Ports With PassiveLoads M. VIDYASAGAR, MEMBER, IEEE Abstracr-The problem of drawing maximum power from an n port by suitably terminating it with another passive n port is studied. Methods are derived for determining sets of passive impedance matrices which are “optimal” in the sense that they draw maximum power from a given n port. It is shown that, in all nontrivial cases, if there exists any (passive or otherwise) optimum-load impedance matrix, then there exist, in fact, infinitely many passive optimum-load impedance matrices. INTRODUCTION T HE maximum power-transfer problem for it ports has been studied by several authors [I], [2]. Consider the network shown in Fig. 1, and suppose the vector of source voltages e and the source impedance matrix Z, are fixed. The problem is to determine a load impedance matrix Z, which has the property that it draws the maximum possible power from e. The best results available to date are due to Desoer [a]. His results are briefly summarized here in the interests of completeness. In order to obtain a simple solution to this problem, we treat i as the independent variable instead of Fig. 1. 2,. Thus, in terms of i, the voltage u is v = e - Z,i while the power delivered (1) to the load is P, = 3 Re (u*i) = $(u*i + i*u) = *(e*i + i*e - i*(Z, + Z,*)i) (2) where (*) denotes conjugate transpose. Since P, is a quadratic form in i, it is very straightforward to characterize “optimal” i. The details can be found in [2]. Case I): Z, + Z,* is positive definite. Let i, = (Z, + Z,*)-‘e. Manuscript received March 12, 1973; revised August 6, 1973. This research was supported by the National Research Council of Canada under Grant A-7790 and by NASA under Grant NGR-05-007-122. This work was performed at the System Science Department, University of California? Los. Angeles, . ^Calif. .1 ne autnor ISon leave from the Department of Electrical Engineering, Sir George Williams University, Montreal, Canada. Any Z resulting in this value i, for i is “optimal” in the sense that it draws maximum power. Note that, by Kirchhoff’s voltage law, we must also have Z,i + Z,i = e. Authorized licensed use limited to: Univ of Texas at Dallas. Downloaded on February 18,2010 at 10:40:10 EST from IEEE Xplore. Restrictions apply. (4) 328 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, MAY 1974 Thus, from (3) and (4), we get a single frequency (say 0). Thus, in conditions a), b), and c), it should be understood that Z and Z, are, in fact, Z,i, = Z,*i, (5) Z(jo) and Z,(jw). In the caseof resistivey1ports, of course, where i, is given by (3), and any Z, satisfying (5) is optimal. conditions a), b), and c) characterizethe maximum powerCase 2): Z, + Z,* is indefinite. No optimal i exists in transfer problem at all frequencies. this case. Secondly,for the purposesof this paper, we define a load Case3~): Z, + Z,* is positive semidefinite,and e belongs impedance matrix Z to be passive if b) holds. Physically, to the range of Z, + Z,*. In this case,the equation this condition ensuresthat if Z is driven by a sinusoidal (vector) current source of frequency w, then the net energy e = (Z, + Z,*)i (6) flow is into Z. Now, as o is varied and the maximum has an infinite number of solutions for i. Any Z, satisfying power-transfer problem is solvedfor all w, then the resulting optimal Z(jw) will satisfy b) for eacho. However, this does Z,i = Z,*i (7) not mean that all poles of Z(s) lie in the closed left halfwhere i is any solution of (6), is optimal. plane. It does not appear to be possible to overcome this Case 3b): Z, + Z,* is positive semidefinite, and e does problem in a simple way. not belong to the range of Z, + Z,*. No optimal i exists in CONSTRUCTIVEPROCEDURES this case. Consider first condition a). The set of all matrices Z in PROBLEMFORMULATION C “xn satisfying a) (for fixed i,) can be characterized very While Desoer’s results are very comprehensive,they do simply. Let IV(&) denote the (n’ - n) dimensional subspace not include any specialconsiderationsabout the passivity or of C”“” consisting of all matrices that map i, into the zero otherwise of the load impedancematrices. From a practical vector. Then, the set S of all Z satisfying a) is given by point of view, it is of interest to draw maximum power from s = {ZE cnxn: Z = Z,* + Z,, Z, E N(i,)>. (8) the given II port using a passiveload, especiallyfrom design considerations. The results of this paper are addressedto The proof of (8) is elementary. First of all, suppose Z the maximum power-transfer problem for IZports when the satisfies a); then, Z, = Z - Z,* belongs to N(i,). On the load impedance matrix is required to be passive, and it is other hand, supposeZ, E IV(&); then if Z = Z,* + Z,, we further required that the network obtained by inter- have Zi, = Z,*i, + Z,i,, = Zs*io, so that Z satisfies a). connecting Z, and Z, be solvable. For our purposes, an Thus the set of all matrices satisfying a) is the linear variety n x n matrix Z, or equivalently an element Z of C” ‘“, is obtained by translating the subspaceN(i,) by the element consideredto be passiveif Z + Z* is positive semidefinite. Z,X. The problem under consideration can be stated as In order for (8) to be useful, we must be able to determine follows: suppose e and Z, are given and that we are in the subspaceN(i,) explicitly. This is easily done. First of Case 1) or Case 3a), so that at least one optimum-load all, note that assumption 2) together with (6), implies that impedancematrix exists. We wish to determine one or more i, # 0. Thus we can normalize i, to be of norm one, or matrices Z in C” ’ ” satisfying the following three conditions : equivalently, we can assumethat jli,ll = 1. Let {io,uz, - . *,v,} be any orthonormal basis for C”. (Such a basis can be a) Zi, = Z,*i, for some solution i, of (6) (optimality); easily generatedusing the Gram-Schmidt procedure.) Let b) Z + Z* is positive semidefinite (passivity); T denote the unitary matrix c) Z + Z, is nonsingular (solvability). We remark that, in Case l), Z = Zs* is an impedance matrix that satisfiesall three conditions, while in Case 3a), 2 = Z,* satisfies a) and b), but not c). In this paper, we prove much stronger results. We make the following rather natural assumptions: 1) n 2 2, i.e., the one port caseis excluded; 2) e # 0, i.e., the source voltages are nontrivial. Subject only to these two assumptions, we show that there exist, in fact, infinitely many matrices in C”“” satisfying conditions a), b), and c). Methods of generating such matrices are also given. Theseresults are particularly useful in Case3a), when the natural choice Z = Z,* is unsuitable because it results in an indeterminate network (since Z, + Z,* is singular). Some comments are in order regarding the given formulation of the problem. First of all, it should be noted that we are tackling the problem of maximum power transfer at * II :.I (9) Then, N(iJ consists of all the matrices of the form TMT*, where M is any matrix whose first column is identically zero. Equivalently, IV(&) is the space generated by the n2 - n matrices 112 0* 77 - 7-. -- 7 v2 3. . . * The proof of this last statement is as follows. Let Z, be a matrix in C” ’ n. Then, the representationof Z, with respect to the basis{iora2,.. . ,u,} is given by the matrix M = T*Z,T. Now the condition that Z,i, = 0 simply means that M must map the vector [lo- . -01’ into the zero vector, or, equivalently, that the first column of M is identically zero. Since Z = TMT*, we seethat N(i,,) consists of all matrices Authorized licensed use limited to: Univ of Texas at Dallas. Downloaded on February 18,2010 at 10:40:10 EST from IEEE Xplore. Restrictions apply. 329 VlLJYASAGAR:POWERTRANSFERIN?ZPORTS of the form TMT*, wherethe first column of M is identically zero. Thus we have identified all matrices satisfying a). Next, consider condition b). Clearly, if 2, and Z, are passive impedance matrices, then so is Z, + Z,. So if Z, is any passive element of N(i,), then Z = Z, + Z,* is a passiveelementof S and, therefore, satisfiesboth conditions a) and b). (Note that Z,* is passive, since we assumethat we are in either Case 1) or Case3a).) Thus we seekpassive elementsof N(i,). Now, supposeZ, = TMT* is an element of N(i,). Then, it is easy to see that Z, is passive, if and only if M is passive.One can think of M as the representation of the matrix operator Z, with respect to the basis { io,vz, * * - ,v,}. Thus let M be any passivematrix whose first column is identically zero. (It is clear that there are infinitely many such matrices. For example, M = diag {O,CI~, * * *,a,,}, where Re CL~2 0 for j =’ 2, * . . ,n. In fact, all such matrices form a convex cone in C”‘“.) Then Z,* + TMT* satisfiesboth a) and b). Consider now condition c). Here we are forced to make a distinction between Cases 1) and 3a). First, suppose Z, + Z,* is positive definite. We can show very easily that there exist infinitely many Z E S satisfying a), b), and c). SupposeM = diag {0,a2; * .,a,} where the aj are all real and aj 2 0 forj = 2, * * *,n. Then, Z = Z,* + TMT* satisfiesa) and b). Further, 2 + Z, = Z, + Z,* + TMT* is the sum of a positive definite matrix and positive semidefinite matrix, whence Z + Z, is itself positive definite, and therefore nonsingular. Hence, Z satisfiesa), b), and c). Alternatively, let M be any passive matrix whose first column is identically zero and consider Z = Z,* + cTMT*, where c is a positive real number. Then, Z + Z, = Z, + Z,* + cTMT*. Since Z, + Z,* is nonsingular, so is Z + Z, for small enough values of c. In fact, [4, p. 5841 Z + Z, is invertible whenever0 I c < 1/(11(Z, + Z,*)-’ 11* l/MI]). Either way, we see that there exist infinitely many impedancematrices satisfying conditions a), b), and c), Now suppose we are in Case 3a), so that Z, + Z,* is singular. In this case,(6) has infinitely many solutions. Let i. denote the solution of (6) with minimum norm, and form the matrix T as before.’ In what follows, we show that, wheneverM is a block diagonal matrix of the form positive definite, it follows that y = 0, where x*Px = PlllXl12 = 0. Sincepi I > 0, lx11 = 0, where x = 0. Q.E.D. Lemma 2: Let i, be the solution of (6) with minimum norm. Then, e*i, > 0. Proof: Assumption 2) implies that i, # 0, as well as that Z, + Z,* # 0. Let {w,, . * . ,w,} denotean orthonormal set of eigenvectors for Z, + Zs*, corresponding to the eigenvalues {i,J2;~~,&,,0; +*,O}, where ~j > 0 for j = I,*** ,m I n - 1. Then, since e belongs to the range of Z, + Zs*, we have (11) where the summation in (11) only goesup to m. Now, i, is given by Hence, i,*e = e*i, = i,*( Z, + Z,*)i, = f Icj12/Aj > 0. j=l (13) Q.E.D. Now, to prove our original claim, let P = T*(Z, + Z,*)T (14) and choose Z = Z,* + TMT*, where M is given by (10). Then, Z + Z, = T(P + M)T*, and Z + Z, is nonsingular, if and only if P + M is nonsingular. However, note that a) P is positive semidefinite; and b) prl = i,*(Z, + Z,*)i, > 0, by Lemma 2. Hence, by Lemma] 1, P + M is positive definite and therefore nonsingular. Thus we seethat Z satisfiesall three conditions a), b), and c). In fact, by proceeding as with Case I), one can easily show that Z = Z,* + T(M + M,)T*, where M is given by (lo), also satisfiesa), b), and c) wheneverthe first column of M/ is identically zero and the norm of M, is small enough. Hence, even in Case3a), there exist infinitely many passive optimal-load impedancematrices. SUMMARY In summary, the procedurefor generatingsetsof optimal passiveimpedancematrices is as follows. where M, is Hermitian and positive definite, the matrix 1) Obtain the solution of (6) with minimum norm (or the Z = Z,* + TMT* satisfies conditions a), b), and c). We unique solution of (6) in caseZ, + Z,* is nonsingular). prove this by means of two lemmas. 2) Form the orthonormal basis {i0,u2,* * . ,u,} and the Lemma I: Let P be a Hermitian positive semidefinite matrix T. matrix with pr I > 0, and let M be as in (10) et seq. Then, 3) Let M be a matrix of the form (lo), and let M, be P + M is positive definite. a matrix whose first column is identically zero. Then, Proof: Consider the triple product x*(P + M)x. We Z = Z,* + T(M + cM,,)T* satisfiesconditions a), b), and wish to show that x*(P + M)x = 0 implies that x = 0. c) for small enough values of the constant c, including Partition x as [x1 1y’]‘. Then, x*(P + M)x = x*Px + c = 0. y*M,y. SinceP + M is at least positive, x*(P + M)x = 0 implies that x*Px = 0 and y*M,y = 0. Since M, is ILLUSTRATIVE EXAMPLES 1 One can easily determine i. using the procedure in [3]. See specifically Remark (4) on p. 216. Example 1 Consider the network in Fig. 1, and let Authorized licensed use limited to: Univ of Texas at Dallas. Downloaded on February 18,2010 at 10:40:10 EST from IEEE Xplore. Restrictions apply. 330 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-21, NO. 7 -1 +j -2-j 2 -1 +j 2-j 7 + 2j -2 . 2-j 10 - 2j -1 [ -2-j Then, one can easily verify that Z, + Z,* is positive definite, and that Z,=& I; i, = ($)[l e= [1 - 1 01’. Z,=$ i, = (+)* [I 1 01’; vg = [O 0 I]‘. Then, the matrix T becomes T = [-ig 2+j -3+j [ .i In this case,one can easily verify that Z, + Z,* is positive semidefinite and that e belongs to the range of Z, + Z,*. Furthermore, the solution of (6) with minimum norm is Choose the vectors v2 and v3 as v2 = (l/J2)[1 :;; ;]. 3, MAY 1974 -1 01’. Let us choose the same T matrix as in Example 1, and let M = diag {0,2a$}, where a > 0, fi > 0 (as opposed to a 2 0, /? 2 0 in Example 1). Then, we seethat any matrix of the form Z, = Z, + TMT” -3-j Let M = diag {0,2a,B}, where a 2 0, j3 2 0. Then, using the constructive procedures given in the previous section, any matrix of the form 2+j l+j l+j 5-2j a a 0 a a 0 I[ 1 -j + 0 0 P is an optimal-load impedancematrix and satisfiesconditions a), b), and c). Z, = Z,* + TMT* 7 a a 0 -1 -j -2+j REFERENCES -1 -j -1-2j 2+j + a a 01 P. M. Lin, “Determinationof availablepower from resistive l11 -2+j 2+j 10 + 2j [ 0 0 P1 multiports,” IEEE Trans. Circuit Theory, vol. CT-19, pp. 385-386, Tnlv 1973 constitutes an optimum-load impedancematrix and satisfies PI C!:Desoer, “The maximum power-transfer theorem for n ports,” IEEE Trans. Circuit Theory, vol. CT-201 pp. 328-329, May 1973. conditions a), b), and c). I31M. Vidyasagar, “Interative minimizatton in Hilbert space of quadratic functtonals with nonunique minima,” Proc. 8th Aflerton Example 2 Co&, pp. 210-218, 1970. I41N. Dunford and J. T. Schwartz, Linear Operntors. Part I. New Consider again the network in Fig. 1, and let York: Interscience, 1959. =A 1 Synthesisof Biconnected Graphs F. T. BOESCH, MEMBER,IEEE, AND J. A. M. MCHUGH Absiruct-In this investigation, it is necessary to distinguish pseudographs (self-loops and multiple lines allowed), multigraph (no self-loops but multiple lines are allowed), and graphs (neither self-loop nor multiple lines allowed). The problem of synthesizing graphs, multigraphs, and pseudographs having a prescribed degree sequence was solved by Hakimi. He also determined the conditions under which a connected realization exists In each of these three cases. The case of biconnected (nonseparable) realizations of a degree sequence was given by Hakimi for the case of pseudographs and multigraphs. His methods did not apply to the case of graphs. The triconnected case was solved by Rao and Rao for graphs; however, strangely enough, the biconnected case apparently remains unpublished. We shall show here that the biconnected case can be handled by a “surgery” technique similar in spirit to the connected case given by Hakimi. Several remarks concerning the general n-connected case for Manuscript received May 8, 1973. The authors are with Bell Telephone Laboratories, Holmdel, NJ pseudographs, multigraphs, and graphs will be given. We conclude with a status report on this subject, which includes the most recent work of Hakimi, Wang and Kleitman, and Bondy. I. INTRODUCTION HE basic notation and terminology follows that of Harary [l]. The problem considered here is the characterization of graphical partitions (degreesequences) of 2-connected graphs. If II = (d,,d,, * * . ,d,,) with di 2 d2 2.e. 2 d,, and p 2 3 is the degreesequenceof some graph, then when does there exist a 2-connectedgraph with this degree sequence?A similar question, as well as the question of the existenceof a connected graph, was solved by Hakimi [2] for the case of multigraphs. Hakimi’s T solution for l-connected multigraphs may be easily modified to hold true for l-connected graphs. However, his proof for Authorized licensed use limited to: Univ of Texas at Dallas. Downloaded on February 18,2010 at 10:40:10 EST from IEEE Xplore. Restrictions apply.