52nd IEEE Conference on Decision and Control

December 10-13, 2013. Florence, Italy

Optimal Power Flow in Direct Current Networks

Lingwen Gan, Steven H. Low

Abstract The optimal power flow (OPF) problem seeks to

control power generation/demand to optimize certain objectives
such as minimizing the generation cost or power loss. Direct
current (DC) networks (e.g., DC-microgrids) are promising to
incorporate distributed generation. This paper focuses on the
OPF problem in DC networks. The OPF problem is nonconvex,
and we study solving it via a second-order cone programming
(SOCP) relaxation. In particular, we prove that the SOCP
relaxation is exact if there are no voltage upper bounds, and
that the SOCP relaxation has at most one solution if it is exact.

30-, 57-, and 118-bus networks [15], though not exact in

general [16]. A through study on the computational speed
and exactness of the SDP relaxation, for some publicly
available networks, can be found in [17].
Sufficient conditions for the exactness of the SDP relaxation have been found for special networks, e.g., (mesh) DC
networks [15], [18], and tree (AC) networks [19][23].
Summary

I. I NTRODUCTION
The optimal power flow (OPF) problem seeks to control
power generation/demand to optimize certain objectives such
as minimizing the generation cost or power loss [1]. It is one
of the fundamental problems in power system operation.
There is increasing interest in direct current (DC) networks, especially DC-microgrids, due to their advantage in
accommodating distributed energy resources such as distributed generation, storage devices, and plug-in electric
vehicles [2]. DC networks are particularly attractive when
loads are DC in nature, e.g., in data centers. This paper
focuses on the OPF problem in DC networks.
The OPF problem is difficult to solve due to the nonconvex
power flow physical lows, and there are in general three ways
to deal with this challenge: (i) linearize the power flow laws;
(ii) look for local optima; and (iii) convexify the power flow
laws, which are described in turn.
Power flow laws can be approximated by some linear
equations in transmission networks, and then the OPF problem reduces to a linear program [3][5]. This method is
widely used in practice and often quite effective, but there
is no guarantee that a feasible solution be found.
Various algorithms have been proposed to find a local optimum of the OPF problem, e.g., successive linear/quadratic
programming [6], trust-region based methods [7], [8], Lagrangian Newton method [9], and interior-point methods
[10][12]. Some of these algorithms, especially NewtonRalphson based, are quite successful empirically. But in
general, there is no guarantee that these algorithms converge,
nor converge to nearly optimal solutions.
Convexification methods are the focus of this paper. It is
proposed in [13][15] to transform the power flow constraints
into linear constraints on a positive semidefinite rank-one
matrix, and then remove the rank-one constraint to obtain a
semidefinite programming (SDP) relaxation. If every solution
of the SDP relaxation is of rank one, then a global optimum
of the OPF problem can be recovered by solving the SDP
relaxation. In this case, the SDP relaxation is called exact.
Strikingly, the SDP relaxation is exact for the IEEE 14-,
978-1-4673-5717-3/13/$31.00 2013 IEEE

The goal of this paper is to propose a convex relaxation for

the OPF problem in (mesh) DC networks, with a sufficient
condition for its exactness. In particular, contributions of this
paper are threefold.
First, we propose a second-order cone programming
(SOCP) relaxation for the OPF problem in (mesh) DC
networks rather than the SDP relaxation. The SOCP relaxation has a much lower computational complexity than the
SDP relaxation, and is exact under existing conditions that
guarantee the exactness of the SDP relaxation. Note that the
SOCP relaxation is coarser than the SDP relaxation [24].
Second, we prove that the SOCP relaxation is exact if
there are no voltage upper bounds. Hence, if voltage upper
bounds are not binding, then the SOCP relaxation is exact.
Third, we prove that the SOCP relaxation has at most one
solution if it is exact.
II. T HE OPTIMAL POWER FLOW PROBLEM
This paper studies the optimal power flow (OPF) problem
in direct current (DC) networks. In the following we present
a model of this scenario that incorporates nonlinear power
flow physical laws, and considers generators/loads as control elements, voltage regulation as control constraints, and
generation cost as control objective.
A. Power flow model
A DC network is composed of buses and lines connecting
these buses. Its topology can be either a tree or a mesh.
There is a swing bus in the network with a fixed voltage.
Index the swing bus by 0 and the other buses by 1, . . . , n.
Let N := {0, . . . , n} denote the collection of all buses and
define N + := N \{0}. Each line connects an unordered pair
{i, j} of buses. Let E denote the collection of all lines and
abbreviate {i, j} E by i j.
For each bus i N , let Vi denote its voltage, Ii denote
its current injection, and pi denote its power injection. For
each line i j, let yij denote its admittance and Iij denote

5614

the current flow from bus i to bus j. In a DC network, Vi ,

Ii , pi , yij , and Iij are all real numbers.1

Bus j
Vj

Bus i
Vi

yij

Ii
pi
Fig. 1.

Summary of the notations.

The notations are summarized in Fig. 1. Further, we use a

letter without subscript to denote a vector of the corresponding quantities, e.g., V = (V1 , . . . , Vn ), y = (yij , i j).
Note that the subscript 0 is not included in nodal variables
to highlight the speciality of the swing bus.
Given the network graph (N , E), the admittance y, and the
swing bus voltage V0 , then the other variables (p, V, I, p0 )
are described by the following physical laws.
1) Ohms Law:
Iij = yij (Vi Vj ),

i j;

distribution networks (DC microgrids are distribution networks), voltages deviate significantly from their nominal
values, and therefore voltage regulation is an important
operational constraint. In distribution networks, lines are usually over-provisioned to support the demand, and therefore
line constraints are not likely to bind. Even if some line
constraints are binding, the utility company can upgrade
the line (distribution lines are much easier and cheaper
to construct than transmission lines). We do not consider
security constraints in distribution networks for simplicity.
Voltage regulation imposes the voltage Vi at bus i to be
maintained within externally specified range [V i , V i ], i.e.,

For example, if 5% voltage deviation from the nominal value

is allowed, then 0.95 Vi 1.05 per unit [25].
The control objective is minimizing the generation cost.
For i N , let fi (pi ) denote the generation cost of bus i
where fi is a real-valued function defined on R. Then, the
generation cost is
X
C(p, p0 ) =
fi (pi ).
(4)

2) Current balance:
Ii =

iN

Iij ,

C. The OPF problem

i N;

j: ji

3) Power balance:
pi = Vi Ii ,

i N.

If we are only interested in voltages and power, then the

three sets of equations can be combined into
X
pi = Vi
(Vi Vj )yij ,
i N,
(1)

The OPF problem seeks to minimize the generation cost

(4), subject to power flow constraints (1), power injection
constraints (2), and voltage constraints (3).
X
OPF: min
fi (pi )
iN

over p, V, p0
s.t. pi = Vi

(Vi Vj )yij ,

pi pi pi ,

which is used to model the power flow in this paper.

V i Vi V i ,

B. Control elements, constraints, and objective

Power generation/demand p of generators/loads can be
controlled, after specifying which the other variables (V, p0 )
are determined by the power flow constraint (1).
The power injection pi of a bus i N + can be controlled
to vary within some externally specified range [pi , pi ], i.e.,
pi pi pi ,

i N;

j: ji

j: ji

i N +.

(2)

For example, if bus i has an inelastic load with power consumption di , and a generator that can generate any amount
of power between 0 and its capacity Ci , then pi = di and
pi = Ci di . Note that pi and pi can be positive or negative.
There are many operational constraints in a practical
OPF problem, including voltage constraints, line (thermal)
constraints, security constraints, etc. In this work, we only
consider voltage constraints for the following reasons. In
1 Voltages,

(3)

i N +.

V i Vi V i ,

currents, admittances, and power are considered real numbers

for DC networks in this paper. While the lines and loads may be inductive
or capacitive during transients at the sub-seconds timescale, they can be
considered as purely resistive at the minutes timescale of the OPF problem.

i N +;
i N +.

The following assumptions are made throughout this work.

A1 The network graph (N , E) is connected.
A2 The swing bus voltage V0 is given, fixed, and non-zero.
A3 The admittances are positive, i.e., yij > 0 for i j.
A4 The bounds V i are positive, i.e., V i > 0 for i N + .
Assumptions A1, A3, and A4 hold in practice. Regarding
Assumption A2, the swing bus voltage V0 can be changed
several times a day, and therefore can be considered as a
constant at the minutes timescale of the OPF problem.
OPF is nonconvex due to the quadratic equality constraints
(1). To overcome this challenge, one can relax OPF to a
convex problem as in [26]. To state the relaxation, define
Wij := Vi Vj ,

i j or i = j,

(5)

and let W := (Wij )ij or i=j denote the collection of all

such Wij . Define a series of 2 2 symmetric matrices


Wii Wij
W {i, j} :=
,
i j & i < j,
Wji Wjj

5615

then OPF can be equivalently formulated as

X
OPF: min
fi (pi )
iN

over p, W, p0
X
s.t. pi =
(Wii Wij )yij ,

i N;

(6a)

j: ji

pi pi pi ,
2
i

V Wii V

i N +;
2
i,

Wij = Wji 0,

(6b)

i N +;

(6c)

i j & i < j;

(6d)

rank(W {i, j}) = 1,

ij &i<j

(6e)

according to Theorem 1, which establishes a bijective map

between the feasible sets of OPF and OPF, that preserves
the objective value. To state the theorem, let FOPF / FOPF
denote the feasible sets, and let C / C 0 denote the objective
functions, of OPF / OPF. Besides, for any feasible point
x = (p, V, p0 ) of OPF, define a map (x) := (p, W, p0 )
where W is defined according to (5).
Theorem 1 For any x FOPF , the point (x) FOPF
and satisfies C(x) = C 0 ((x)). Furthermore, the map :
FOPF FOPF is bijective.
The theorem is proved in Appendix A.
After transforming OPF to OPF, a second-order cone
programming (SOCP) relaxation can be proposed.
X
SOCP: min
fi (pi )
iN

over p, W, p0
s.t. (6a), (6b), (6c), (6d);
W {i, j}  0,

i j & i < j.

(7)

The SOCP relaxation is proposed in [26] to convexify OPF

for tree networks. In this paper, we use it to convexify OPF
for DC networks, which can be either mesh or tree.
If a solution w = (p, W, p0 ) of SOCP is feasible for OPF,
i.e., w satisfies (6e), then w is a global optimum of OPF.
This motivates a definition of exactness for SOCP.
Definition 1 SOCP is exact if every of its solutions satisfies
(6e).
If SOCP is exact, then a global optimum of OPF can be
easily recovered. More specifically, if SOCP is exact and
(p, W, p0 ) is an optimal solution of SOCP, let Vi = Wii
for i N , then (p, V, p0 ) is a global optimum of OPF.
D. Related work
A semidefinite programming (SDP) relaxation has also
been proposed in literature to convexify OPF [13][15]. The
SDP relaxation enlarges the feasible set of OPF to a smaller
convex set than that of SOCP, and is therefore more likely
to be exact [24]. We propose SOCP rather than the SDP
relaxation for the following reasons:

R1 SOCP has a significantly lower computational complexity than the SDP relaxation;
R2 SOCP is exact under existing conditions that guarantee
the exactness of the SDP relaxation.
To demonstrate R2, we review existing conditions that
guarantee the exactness of both relaxations in Propositions
1 and 2, which follow directly from a more general result in
[27, Theorem 3.1].
Proposition 1 ( [15]) If pi = for i N + and fi is
strictly increasing for i N , then the SDP relaxation is
exact.
Proposition 2 ( [18]) If pi = for i N + and fi is
strictly increasing for i N , then SOCP is exact.
The conditions in Propositions 1 and 2 are the same, which
demonstrates R2.
III. A SUFFICIENT CONDITION
We study the exactness of SOCP in this section. In
particular, a sufficient condition is provided in the following
theorem, which is proved in Section III-A.
Theorem 2 If V i = for i N + and f0 is strictly
increasing, then SOCP is exact.
The theorem implies that if voltage upper bounds are not
binding, then SOCP is exact.
Theorem 2 still holds if power injection constraints (2) are
generalized to p P where P is an arbitrary set, since the
proof of Theorem 2 does not require any structure on P.
Theorem 3 If SOCP is convex and exact, then it has at most
one solution.
The theorem is proved in Appendix B.
Theorem 3 still holds if power injection constraints (2) are
generalized to p P where P is an arbitrary convex set.
A. Proof of Theorem 2
The idea of the proof is as follows. Under the conditions
in Theorem 2, for any feasible point w of SOCP that violates
(6e), there exists another feasible point w0 of SOCP that has
a smaller objective value than w. Hence, every solution of
SOCP must satisfy (6e), i.e., SOCP is exact.
More specifically, Theorem 2 follows from the following
two lemmas, which are proved in Appendix C and D.
Lemma 1 Let w = (p, W, p0 ) be an arbitrary feasible point
of SOCP. If V k = for k N + , and w violates (6e)
on some i j where i, j N + , then there exists another
feasible point w0 = (p, W 0 , p0 ) of SOCP that violates (6e)
on k l whenever {k, l} {i, j} =
6 .
Lemma 1 implies that violation of (6e) propagates to neighboring lines: for any feasible point w of SOCP that violates
(6e) on some line i j, there exists another feasible point

5616

w0 with the same power injection, that violates (6e) on all

neighboring lines of i j (including i j).
Lemma 2 If V k = for k N + and f0 is strictly
increasing, then every solution of SOCP must satisfy (6e)
on all neighboring lines i 0 of the swing bus.
Lemma 2 implies that every solution w of SOCP satisfies
(6e) on all neighboring lines of bus 0.
Combining Lemma 1 and 2 gives the proof of Theorem
2. If there exists a solution w of SOCP that violates (6e) on
some line, then since the network is connected, by repeating
the propagation procedure described in Lemma 1, one can
find a feasible point w0 of SOCP with the same power
injection as w, that violates (6e) on some neighboring line
k 0 of the swing bus 0.
The feasible point w0 has the same objective value as w
and is therefore optimal. This contradicts Lemma 2. Hence,
every solution of SOCP must satisfy (6e), i.e., SOCP is exact.
This completes the proof of Theorem 2.
IV. C ONCLUSION
We have proposed an SOCP relaxation for the OPF
problem in DC networks, with a sufficient condition under
which the SOCP relaxation is exact. Specifically, we have
proved that the SOCP relaxation is exact if voltage upper
bounds are not binding. We have also proved that the SOCP
relaxation has at most one solution if it is exact.
There remains many interesting open questions. For example, what happens if we consider line constraints? If the
SOCP relaxation is not exact, can a nearly optimal solution
of the OPF problem be recovered?
R EFERENCES
[1] J. Carpentier, Contribution to the economic dispatch problem,
Bulletin de la Societe Francoise des Electriciens, vol. 3, no. 8, pp.
431447, 1962.
[2] H. Kakigano, Y. Miura, and T. Ise, Low-voltage bipolar-type dc
microgrid for super high quality distribution, IEEE Transactions on
Power Electronics, vol. 25, no. 12, pp. 30663075, 2010.
[3] B. Stott and O. Alsac, Fast decoupled load flow, IEEE Transactions
on Power Apparatus and Systems, vol. PAS-93, no. 3, pp. 859869,
1974.
[4] O. Alsac, J. Bright, M. Prais, and B. Stott, Further developments in
lp-based optimal power flow, IEEE Transactions on Power Systems,
vol. 5, no. 3, pp. 697711, 1990.
[5] B. Stott, J. Jardim, and O. Alsac, Dc power flow revisited, IEEE
Transactions on Power Systems, vol. 24, no. 3, pp. 12901300, 2009.
[6] G. C. Contaxis, C. Delkis, and G. Korres, Decoupled optimal power
flow using linear or quadratic programming, IEEE Transactions on
Power Systems, vol. 1, no. 2, pp. 17, 1986.
[7] W. Min and L. Shengsong, A trust region interior point algorithm
for optimal power flow problems, International Journal on Electrical
Power and Energy Systems, vol. 27, no. 4, pp. 293300, 2005.
[8] A. A. Sousa and G. L. Torres, Robust optimal power flow solution
using trust region and interior methods, IEEE Transactions on Power
Systems, vol. 26, no. 2, pp. 487499, 2011.
[9] E. C. Baptista, E. A. Belati, and G. R. M. da Costa, Logarithmic
barrier-augmented lagrangian function to the optimal power flow problem, International Journal on Electrical Power and Energy Systems,
vol. 27, no. 7, pp. 528532, 2005.
[10] G. L. Torres and V. H. Quintana, An interior-point method for
nonlinear optimal power flow using voltage rectangular coordinates,
IEEE Transactions on Power Systems, vol. 13, no. 4, pp. 12111218,
1998.

[11] R. A. Jabr, A primal-dual interior-point method to solve the optimal

power flow dispatching problem, Optimization and Engineering,
vol. 4, no. 4, pp. 309336, 2003.
[12] F. Capitanescu, M. Glavic, D. Ernst, and L. Wehenkel, Interior-point
based algorithms for the solution of optimal power flow problems,
Electric Power System Research, vol. 77, no. 5-6, pp. 508517, 2007.
[13] X. Bai, H. Wei, K. Fujisawa, and Y. Yang, Semidefinite programming
for optimal power flow problems, International Journal of Electric
Power & Energy Systems, vol. 30, no. 6, pp. 383392, 2008.
[14] X. Bai and H. Wei, Semi-definite programming-based method for
security-constrained unit commitment with operational and optimal
power flow constraints, Generation, Transmission & Distribution,
IET,, vol. 3, no. 2, pp. 182197, 2009.
[15] J. Lavaei and S. H. Low, Zero duality gap in optimal power flow
problem, IEEE Transactions on Power Systems, vol. 27, no. 1, pp.
92107, 2012.
[16] B. Lesieutre, D. Molzahn, A. Borden, and C. L. DeMarco, Examining
the limits of the application of semidefinite programming to power
flow problems, in Allerton, 2011, pp. 14921499.
[17] D. K. Molzahn, J. T. Holzer, B. C. Lesieutre, and C. L. DeMarco,
Implementation of a large-scale optimal power flow solver based on
semidefinite programming, IEEE Transactions on Power Systems.
[18] J. Lavaei, A. Rantzer, and S. Low, Power flow optimization using
positive quadratic programming, in Proceedings of IFAC World
Congress, 2011.
[19] B. Zhang and D. Tse, Geometry of injection regions of power
networks, IEEE Transactions on Power Systems, to appear, 2012.
[20] S. Bose, D. Gayme, S. Low, and K. Chandy, Quadratically constrained quadratic programs on acyclic graphs with application to
power flow, arXiv:1203.5599, 2012.
[21] M. Farivar, C. R. Clarke, S. H. Low, and K. M. Chandy, Inverter var control for distribution systems with renewables, in IEEE
SmartGridComm, 2011, pp. 457462.
[22] L. Gan, N. Li, U. Topcu, and S. H. Low, On the exactness of convex
relaxation for optimal power flow in tree networks, in IEEE CDC,
2012, pp. 465471.
[23] A. Lam, B. Zhang, A. Dominguez-Garcia, and D. Tse, Optimal
distributed voltage regulation in power distribution networks, arXiv
preprint arXiv:1204.5226, 2012.
[24] S. Bose, S. Low, and K. Chandy, Equivalence of branch flow and
bus injection models, in Proc. of 50th Annual Allerton Conference,
October 2012.
[25] Electric power systems and equipmentvoltage ratings (60 hertz),
ANSI Standard Publication, no. ANSI C84.1, 1995.
[26] S. Sojoudi and J. Lavaei, Physics of power networks makes hard
optimization problems easy to solve, in IEEE Power & Energy
Society General Meeting, 2012, pp. 18.
[27] S. Kim and M. Kojima, Exact solutions of some nonconvex quadratic
optimization problems via sdp and socp relaxations, Computational
Optimization and Applications, vol. 26, no. 2, pp. 143154, 2003.

A PPENDIX
A. Proof of Theorem 1
It is straightforward to show that, for any feasible point
x = (p, V, p0 ) of OPF, if one defines W according to (5),
then the point (x) = (p, W, p0 ) is feasible for OPF and
has the same objective value as x. It remains to show that
the map is bijective, i.e., both injective and surjective.
First show that the map is injective, i.e., for x, x0
FOPF , if (x) = (x0 ), then x = x0 . Denote x = (p, V, p0 )
and x0 = (p0 , V 0 , p00 ), then (x) = (p, W, p0 ) where W
satisfies Wij = Vi Vj for i j or i = j, and (x0 ) =
(p0 , W 0 , p00 ) where W 0 satisfies Wij0 = Vi0 Vj0 for i j or
i = j. Therefore it follows from (x) = (x0 ) immediately
that p = p0 and p0 = p00 . To show that x = x0 , it remains to
prove that V = V 0 .
It follows from (x) = (x0 ) that W = W 0 , which implies
Wii = Wii0 for i N , i.e.,

5617

Vi2 = Vi02 ,

i N.

Since Vi V i > 0 and Vi0 V i > 0, one has Vi = Vi0 for

i N . This completes the proof that the map is injective.
Finally show that the map is surjective, i.e., for any
(p, W, p0 ) FOPF , there
exists x FOPF such that (x) =
(p, W, p0 ). Set Vi = Wii for i N and define x :=
(p, V, p0 ). It remains to show that (x) = (p, W, p0 ) and x
is feasible for OPF.
To show that (x) = (p, W, p0 ), it suffices to show that
W
ij
= Vi Vj for i j or i = j. It follows from Vi =
Wii immediately that Wij = Vi Vj for
pi = j. Noting that
rank(W {i, j}) = 1, one has Wij = Wii Wjj = Vi Vj
for i j. Since Wij 0, it must be that Wij = Vi Vj for
i j. Hence, (x) = (p, W, p0 ).
Having established the relationship that Wij = Vi Vj for
i j or i = j, it is straightforward to show that x is feasible
for OPF.
To this end, it has been proved that the map is bijective.
This completes the proof of Theorem 1.
B. Proof of Theorem 3
Assume that SOCP is convex, exact, and has at least one
, p0 ) and w
, p0 ) be two
solution. Let w
= (
p, W
= (
p, W
arbitrary solutions of SOCP, then it suffices to prove that
w
= w.

Define w := (w
+ w)/2,

then w is feasible and optimal

for SOCP since SOCP is convex. Denote w = (p, W, p0 ). It
follows from SOCP being exact that w,
w,
and w all satisfy
(6e). Hence,
ii W
jj = W
ij2 ,
W
ii W
jj = W
2,
W
ij
Wii Wjj =

(8a)
(8b)
(8c)

Wij2

C. Proof of Lemma 1
It suffices to prove that, given any feasible point w =
(p, W, p0 ) of SOCP that violates (6e) on some line i
j where i, j N + , there exists a feasible point w0 =
(p0 , W 0 , p00 ) of SOCP with (p0 , p00 ) = (p, p0 ), that violates
(6e) on k l whenever {k, l} {i, j} =
6 .
Note that (6e) is equivalent to
p
k l,
Wkl = Wkk Wll ,
and (7) is equivalent to
0 Wkl

i j.

q
ii W
jj W
ii W
jj
2 W
ii W
jj + W
ii W
jj
W

After constructing W 0 , construct (p0 , p00 ) as

X
0
0
p0k =
(Wkk
Wkl
)ykl ,
k N.

(10)

l: lk

It remains to prove that the point w0 = (p0 , W 0 , p00 ) is feasible

0 0
for SOCP,
, p0 ) = (p, p0 ), and w0 violates (6e) (i.e.,
p (p
0
0
6 .
Wkl 6= Wkk Wll0 ) on k l whenever {k, l} {i, j} =
First prove that (p0 , p00 ) = (p, p0 ). When k
/ {i, j},
X
0
0
p0k =
(Wkk
Wkl
)ykl
l: lk

=
=

(Wkk Wkl )ykl

l: lk
pk .

l: lk
0
Wkk

Wkk

ykl

l: lk

for i j. The inequality attains equality, which implies

ii W
jj = W
ii W
jj for i j. Since W
kk , W
kk V 2 > 0
W
k
+
2

for k N and W00 = W00 = V0 6= 0, one has

ii
jj
W
W
=
,
ii
jj
W
W

k l.

When k {i, j}, one has

X
0
0
p0k =
(Wkk
Wkl
)ykl

It follows that
ii W
jj + W
ii W
jj
W

Wkk Wll ,

If w violates (6e) on i j but satisfies (7), then

p
Wij < Wii Wjj .
p
Pick an  such that 0 <  <
Wii Wjj Wij , and
construct W 0 as

if {k, l} = {i, j}

Wkl + 
y
ij
0
 if {k, l} = {i} or {j} (9)
Wkl = Wkl + P
h: hk ykh

Wkl
otherwise.

+W
)/2 in (8c), and simplify
for i j. Substitute W = (W
using (8a) and (8b) to obtain
ii W
jj + W
ii W
jj = 2W
ij W
ij ,
W

X
X

0
Wkl
ykl

l: lk

ykl + P

yij

l: lk

l: lk

ykl

ykl

l: lk

Wkl ykl yij 

l: lk

i j.

ii /W
ii for i N , then 0 = 1 and i = j
Define i := W
if i j. Since the network (N , E) is connected, it follows
ii = W
ii for i N . Then, it
that i = 1 for i N , i.e., W
follows from (6e) that
q
q
ij = W
ii W
jj = W
ii W
jj = W
ij
W
= W
. It
for i j. To this point, we have shown that W
follows immediately that w
= w,
which completes the proof
of Theorem 3.

Wkk

X
l: lk

=
=

ykl

Wkl ykl

l: lk

(Wkk Wkl )ykl

l: lk
pk .

Hence, (p0 , p00 ) = (p, p0 ).

p 0
0
Next prove that w0 satisfies Wkl
6= Wkk
Wll0 on k l
whenever {k, l} {i, j} 6= . Let k l be an arbitrary line
that satisfies {k, l} {i, j} =
6 . If {k, l} = {i, j}, then
q
p
0
0 W0 .
Wkl
= Wkl +  = Wij +  < Wii Wjj Wkk
ll

5618

First prove that p0 = p and p00 < p0 . When k

/ {0, i},
X
0
0
p0k =
(Wkk
Wkl
)ykl

If {k, l} =
6 {i, j}, then one of k, l is in {i, j}, and the other
is not. Without loss of generality, assume that k = i and
l
/ {i, j}. Then,
q
q
p
0
0 W =
0 W0 .
Wkk
|Wkl
| = |Wkl | Wkk Wll < Wkk
ll
ll
p 0
0
Hence, Wkl
6=
Wkk Wll0 on k l whenever {k, l}
{i, j} =
6 .
Finally prove that w0 is feasible for SOCP if V k = for
k N + . The point w0 satisfies (6a) according to (10). Since
w = (p, W, p0 ) is feasible for SOCP, one has pk pk pk
for k N + . Then it follows from p0 = p that pk p0k pk
for k N + , i.e., w0 satisfies (6b). It follows from (9) that
0
Wkk
Wkk V 2k for k N + , i.e., w0 satisfies (6c) if
V k = for k N + . To check (7), note that for k l, if
{k, l} =
6 {i, j}, then

h p
i  q
0
0
0
Wkl = Wkl 0, Wkk Wll 0, Wkk Wll ;

l: lk

=
=

(Wkk Wkl )ykl

l: lk
pk .

When k = i, one has

X
0
0
p0k =
(Wkk
Wkl
)ykl
=

l: lk
0
Wkk

Wkk

ykl

l: lk

0
Wkl
ykl

l: lk

ykl + P

y0i

l: lk

l: lk

ykl

ykl

l: lk

Wkl ykl y0i 

l: lk

Wkk

=
=

ykl

l: lk

if {k, l} = {i, j}, then


h p
  q
0
0 W0 .
Wkl
= Wkl +  , Wkk Wll 0, Wkk
ll

Wkl ykl

l: lk

(Wkk Wkl )ykl

l: lk
pk .

When k = 0, one has

X
0
0
p0k =
(Wkk
Wkl
)ykl

This completes the proof of Lemma 1.

D. Proof of Lemma 2

It suffices to prove that, given any feasible point w =

(p, W, p0 ) of SOCP that violates (6e) on some line i 0,
there exists a feasible point w0 = (p0 , W 0 , p00 ) of SOCP with
p0 = p and p00 < p0 . Since then,
X
X
fi (p0i )
fi (pi ) = f0 (p00 ) f0 (p0 ) < 0,
iN

l: lk
0
Wkk

ykl

l: lk

Wkk

0
Wkl
ykl

l: lk

ykl

l: lk

X
X

Wkl ykl y0i 

l: lk

(Wkk Wkl )ykl y0i 

l: lk

iN

= pk y0i  < pk .

i.e., the feasible point w has a smaller objective value than

w. It follows that every solution of SOCP must satisfy (6e)
on every neighboring line i 0 of the swing bus 0, which
completes the proof of Lemma 2.

Note that (6e) is equivalent to Wkl

= Wkk Wll for k l,
and (7) is equivalent to 0 Wkl Wkk Wll for k l. If
w violates (6e) on i 0 but satisfies (7), then
p
W0i < W00 Wii .

Pick an  such that 0 <  < W00 Wii W0i , and

construct W 0 as

if {k, l} = {0, i}

Wkl + 
y0i
0
P
W
+

if {k, l} = {i}
Wkl =
(11)
kl
h: hk ykh

Wkl
otherwise.
0

After constructing W 0 , construct (p0 , p00 ) as

X
0
0
p0k =
(Wkk
Wkl
)ykl ,
k N.

Hence, p0 = p and p00 < p0 .

Next prove that w0 is feasible for SOCP if V k = for
k N + . The point w0 satisfies (6a) according to (12). Since
w = (p, W, p0 ) is feasible for SOCP, one has pk pk pk
for k N + . Then it follows from p0 = p that pk p0k pk
for k N + , i.e., w0 satisfies (6b). It follows from (11) that
0
Wkk
Wkk V 2k for k N + , i.e., w0 satisfies (6c) if
V k = for k N + . To check (7), note that for k l, if
{k, l} =
6 {0, i}, then

h p
i  q
0
0 W0 ;
Wkl
= Wkl 0, Wkk Wll 0, Wkk
ll
if {k, l} = {0, i}, then

h p
  q
0
0 W0 .
Wkl
= Wkl +  , Wkk Wll 0, Wkk
ll
This completes the proof of Lemma 2.

(12)

l: lk

It remains to prove that w0 is feasible for SOCP, p0 = p, and

p00 < p0 .
5619