TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS 15

Convex Relaxation of Optimal Power Flow—Part I:

Formulations and Equivalence
Steven H. Low, Fellow, IEEE

Abstract—This tutorial summarizes recent advances in the convex SOCP for radial networks in the branch flow model of [45]. See
relaxation of the optimal power flow (OPF) problem, focusing on Remark 6 below for more details. While these convex relaxations
structural properties rather than algorithms. Part I presents two have been illustrated numerically in [22] and [23], whether or
power flow models, formulates OPF and their relaxations in each
model, and proves equivalence relationships among them. Part II when they will turn out to be exact is first studied in [24].
presents sufficient conditions under which the convex relaxations are Exploiting graph sparsity to simplify the SDP relaxation of OPF
exact. is first proposed in [25] and [26] and analyzed in [27] and [28].
Index Terms—Convex relaxation, optimal power flow, power Convex relaxation of quadratic programs has been applied to
systems, quadratically constrained quadratic program (QCQP), many engineering problems; see, e.g., [29]. There is a rich theory
second-order cone program (SOCP), semidefinite program and extensive empirical experiences. Compared with other
(SDP), semidefinite relaxation. approaches, solving OPF through convex relaxation offers
several advantages. First, while DC OPF is useful in a wide
I. INTRODUCTION variety of applications, it is not applicable in other applications;
see Remark 10. Second, a solution of DC OPF may not be
OR our purposes, an optimal power flow (OPF) problem is
F a mathematical program that seeks to minimize a certain
function, such as total power loss, generation cost or user
feasible (may not satisfy the nonlinear power flow equations). In
this case, an operator may tighten some constraints in DC OPF
and solve again. This may not only reduce efficiency but also
disutility, subject to the Kirchhoff’s laws, as well as capacity,
relies on heuristics that are hard to scale to larger systems or faster
stability, and security constraints. OPF is fundamental in power
control in the future. Third, when they converge, most nonlinear
system operations as it underlies many applications such as
algorithms compute a local optimal usually without assurance on
economic dispatch, unit commitment, state estimation, stability
the quality of the solution. In contrast, a convex relaxation
and reliability assessment, volt/var control, demand response, etc.
provides for the first time the ability to check whether a solution
There has been a great deal of research on OPF since Carpentier’s
is globally optimal. If it is not, the solution provides a lower
first formulation in 1962 [2]. An early solution appears in [3] and
bound on the minimum cost and hence a bound on how far any
extensive surveys can be found in, e.g., [4]–[15].
feasible solution is from optimality. Unlike approximations, if a
Power flow equations are quadratic and hence OPF can be
relaxed problem is infeasible, it is a certificate that the original
formulated as a quadratically constrained quadratic program
OPF is infeasible.
(QCQP). It is generally nonconvex and hence NP-hard. A large
This two-part tutorial explains the main theoretical results on
number of optimization algorithms and relaxations have been
semidefinite relaxations of OPF developed in the last few years.
proposed. A popular approximation is a linear program, called
Part I presents two power flow models that are useful in different
DC OPF, obtained through the linearization of the power flow
situations, formulates OPF and its convex relaxations in each
equations, e.g., [16]–[20]. See also [21] for a more accurate linear
model, and clarifies their relationship. Part II [30] presents
approximation. To the best of our knowledge, solving OPF
sufficient conditions that guarantee the relaxations are exact,
through semidefinite relaxation is first proposed in [22] as a
i.e., when one can recover a globally optimal solution of OPF
second-order cone program (SOCP) for radial (tree) networks
from an optimal solution of its relaxations. We focus on basic
and in [23] as a semidefinite program (SDP) for general networks
results using the simplest OPF formulation and does not cover
in a bus injection model. It is first proposed in [51], [57] as an
many relevant works in the literature, such as stochastic OPF,
e.g., [31]–[33]; distributed OPF, e.g., [34]–[39]; new applica-
Manuscript received September 23, 2013; revised February 5, 2014; tions, e.g., [40], [41]; or what to do when relaxation fails, e.g.,
February 19, 2014; accepted February 28, 2014. Date of publication March 5, [42]–[44], to name just a few.
2014; date of current version April 9, 2014. This work was supported in part by
NSF under Grant NetSE CNS 0911041, in part by ARPA-E under Grant GENI
DE-AR0000226, in part by the National Science Council of Taiwan under Grant A. Outline of Paper
NSC 103-3113-P-008-001, in part by Southern California Edison, in part by the
Los Alamos National Lab, and in part by Caltech’s Resnick Institute. A preliminary Many mathematical models have been used to model power
and abridged version has appeared in [S. H. Low, “Convex relaxation of optimal networks. In Part I of this two-part paper, we present two such
power flow: A tutorial,” in Proc. IREP Symp. Bulk Power Syst. Dyn. Control (IREP),
Rethymnon, Greece, Aug. 2013.]. Recommended by Associate Editor M. Chertkov. models, we call the bus injection model (BIM) and the branch
The author is with Engineering and Applied Science (EAS), California Institute flow model (BFM). Each model consists of a set of power flow
of Technology (Caltech), Pasadena, CA 91125 USA (e-mail: [email protected]). equations. Each models a power network in that the solutions of
Color versions of one or more of the figures in this paper are available online at
http://ieeexplore.ieee.org. each set of equations, called the power flow solutions, describe
Digital Object Identifier 10.1109/TCNS.2014.2309732 the steady state of the network. We prove that these two models

2325-5870 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution
requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
16 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

are equivalent in the sense that there is a bijection between their to denote a directed graph. By “ ,” we mean an
solution sets (Section II). We formulate OPF within each model, edge if is undirected and either or if is
where the power flow solutions define the feasible set of OPF directed. Sometimes we write or to mean
(Section III). Even though BIM and BFM are equivalent some or , respectively. A cycle is an
results are much easier to formulate or prove in one model than ordered set of nodes so that for
the other; see Remark 2 in Section II. with the understanding that . In that
The complexity of OPF formulated here lies in the noncon- case, we refer to a link or a node in the cycle by or
vexity of power flow equations that gives rise to a nonconvex , respectively.
feasible set of OPF. We develop various characterizations of the
feasible set and design convex supersets based on these char-
II. POWER FLOW MODELS
acterizations. Different designs lead to different convex relaxa-
tions and we prove their relationship (Sections IV and V). When In this section, we describe two mathematical models of power
a relaxation is exact, an optimal solution of the original non- networks and prove their equivalence. By a “mathematical
convex OPF can be recovered from any optimal solution of the model,” we mean a set of variables and a set of equations relating
relaxation. In Part II [30], we present sufficient conditions that these variables. These equations are motivated by the physical
guarantee the exactness of convex relaxations. system, but mathematically, they are the starting points from
BFMs are originally proposed for networks with a tree topol- which all claims are derived.
ogy, called radial networks, e.g., [45]–[53]. They take a recur-
sive structure that simples the computation of power flow A. Bus Injection Model
solutions, e.g., [54], [55], [47]. The model of [45] and [46] also
Consider a power network modeled by a connected undirected
has a linearization that offers several advantages over DC OPF in
graph where , ,
BIM; see Remark 10. The linear approximation provides simple
and . Each node in represents a bus and each
bounds on the branch powers and voltage magnitudes in the
edge in represents a transmission or distribution line. We use
nonlinear BFM (Section VI). These bounds are used in [56] to
“bus” and “node” interchangeably and “line” and “edge” inter-
prove a sufficient condition for exact relaxation.
changeably. For each edge , let C be its admit-
Finally, we make algorithmic recommendations in Section VII
tance. A bus can have a generator, a load, both or neither.
based on the key results presented here. Most proofs can be found
Let be the complex voltage at bus and denote its
in the original papers (or the arXiv version of this tutorial) and are
magnitude. Bus 0 is the slack bus. Its voltage is fixed, and we
omitted.
assume without loss of generality that per unit (p.u.).
B. Notations Let be the net complex power injection (generation minus
load) at bus .
Let C denote the set of complex numbers and R the set of real
The BIM is defined by the following power flow equations that
numbers. For C, and denote the real and imagi-
describe the Kirchhoff’s laws:
nary parts of , respectively. For any set C ,
denotes the convex hull of . For R, .
For C, means and . We
abuse notation to use the same symbol to denote either a
complex number or a two-dimensional real vector Let the set of power flow solutions for each be
, ) depending on the context.
In general, scalar or vector variables are in small letters, e.g., V C
. Most power system quantities, however, are in
For convenience, we include in the vector variable
capital letters, e.g., . A variable without a
with the understanding that is
subscript denotes a vector with appropriate components, e.g.,
fixed.
and . For vectors
, denotes component-wise inequality. Remark 1 (Bus Types): Each bus is characterized by two
Matrices are usually in capital letters. The transpose of a complex variables and , or equivalently, four real variables.
matrix is denoted by and its Hermitian (complex conju- The buses are usually classified into three types, depending on
gate) transpose by . A matrix is Hermitian if . which two of the four real variables are specified. For the slack
is positive semidefinite (or psd), denoted by , if is bus 0, is given and is variable. For a generator bus (also
Hermitian and for all C ; in particular, if , called -bus), and are specified and
then by definition, . For matrices means and are variables. For a load bus (also called
is psd. Let S be the set of all Hermitian matrices -bus), is specified and is variable. The power flow or
and S the set of psd matrices. load flow problem is: given two of the four real variables
A graph consists of a set of nodes and a set specified for each bus, solve the complex equations in
of edges. If is undirected then if and (1) for the remaining real variables. For instance, when
only if . If is directed then only if all buses are all load buses, the power flow problem
; in this case, we will use and interchange- solves (1) for the complex voltages , and the power
ably to denote an edge pointing from to . We sometimes use injection at the slack bus 0. This can model a distribution
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 17

system with a substation at bus 0 and constant-power loads at allows a more efficient computation of power flows and leads to a
the other buses. For OPF problems, and on generator useful linear approximation of BFM; see Section VI. The
buses or on load buses can be variables as well. For instance, sufficient condition for exact relaxation in [56] provides
economic dispatch optimizes real power generations at intricate insights on power flows that are hard to formulate or
generator buses; demand response optimizes demands at prove in BIM. Finally, since BFM directly models branch flows
load buses; and volt/var control optimizes reactive powers and currents , it is easier to use for some applications.
at capacitor banks, tap changers, or inverters. These remarks also Therefore, we will freely use either model depending on which is
apply to the BFM presented next. more convenient for the problem at hand.

B. Branch Flow Model III. OPTIMAL POWER FLOW

In the BFM, we adopt a connected directed graph A. Bus Injection Model
where each node in represents
a bus and each edge in represents a transmission As mentioned in Remark 1, an OPF problem optimizes both
or distribution line. Fix an arbitrary orientation for and let variables and over the solution set of the BIM (1). In addition,
be the number of directed edges in . Denote an edge all voltage magnitudes must satisfy
by or if it points from node to node . For each edge
, let be the complex impedance on the
line; let be the complex current and be the
where and are given lower and upper bounds on voltage
sending-end complex power from buses to . For each bus
magnitudes. Throughout this paper, we assume > to avoid
, let be the complex voltage at bus . Assume without
triviality. The power injections are also constrained
loss of generality that p.u. Let be the net complex
power injection at bus .
The BFM in [57] is defined by the following set of power flow
equations: where and are given bounds on the injections at buses .
Remark 3 (OPF Constraints): If there is no bound on the load
or on the generation at bus then or
, respectively. On the other hand, (4) also
allows the case where is fixed (e.g., a constant-power
load), by setting to the specified value. For the slack
where (2b) is the Ohm’s law, (2c) defines branch power, and (2a) bus 0, unless otherwise specified, we always assume
imposes power balance at each bus. The quantity and and . There-
represents line loss so that is the receiving-end fore, we sometimes replace in (3) and (4) by .
complex power at bus from bus . We can eliminate the variables from the OPF formulation by
Let the set of solutions of BFM for each be combining (1) and (4) into

X C

For convenience, we include in the vector variable

with the understanding that is Then, OPF in the BIM can be defined just in terms of complex
fixed. voltage vector . Define

C. Equivalence V C
Even though the BIM (1) and the BFM (2) are defined by
V is the feasible set of OPF problems in BIM.
different sets of equations in terms of their own variables, both
Let the cost function be . Typical costs include the cost of
are models of the Kirchhoff’s laws and therefore must be related.
generating real power at each generator bus or line loss over the
We now clarify the precise sense in which these two mathemati-
network. All these costs can be expressed as functions of .
cal models are equivalent. We say two sets and are
Then, the problem of interest is
equivalent, denoted by , if there is a bijection between
OPF:
them [58].
V
Theorem 1: V X for any power injections .
Remark 2 (Two Models): Given the bijection between the Since (5) is quadratic, V is generally a nonconvex set. OPF is
solution sets V and X any result in one model is in principle thus a nonconvex problem and NP-hard to solve in general.
derivable in the other. Some results, however, are much easier to
state or derive in one model than the other. For instance, BIM B. Branch Flow Model
allows a much cleaner formulation of the SDP relaxation. BFM Denote the variables in the BFM (2) by
for radial networks has a convenient recursive structure that C . We can also eliminate the variables as for the BIM
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
18 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

by combining (2a) and (4) but it will prove convenient to retain where is the -dimensional vector with 1 in the th entry
as part of the variables. Define the feasible set and 0 elsewhere. Hence, since , we have
in the BFM

X C

Let the cost function in the BFM be . Then the OPF

problem in the BFM is
OPF:
where is an matrix with its th
X row equal to the th row of the admittance matrix and all other
rows equal to the zero vector. is in general not Hermitian so
Since (2) is quadratic, X is generally a nonconvex set. OPF is that is in general a complex number. Its real and
thus a nonconvex problem and NP-hard to solve in general. imaginary parts can be expressed in terms of Hermitian and
Remark 4 (OPF Equivalence): By Theorem 1, there is a skew Hermitian components of defined as
bijection between V and X. Throughout this paper, we
assume that the cost functions in BIM and BFM are
equivalent under this bijection and we abuse notation to
denote them by the same symbol . Then, OPF (7) in BIM Then
and (9) in BFM are equivalent.
Remark 5 (OPF Variants): OPF as defined in (7) and (9) is a
simplified version that ignores other important constraints such as
line limits, security constraints, stability constraints, and chance Let their upper and lower bounds be denoted by
constraints; see extensive surveys in [4]–[15], [59], and [60] and
a recent discussion in [61] on real-life OPF problems. Some of
these can be incorporated without any change to the results in this
paper (e.g., see [57] and [62] for models that include shunt
elements and line limits). Indeed a shunt element at bus
can be easily included in BIM by modifying (1) into Let denote the Hermitian matrix with a single 1 in
the th entry and 0 everywhere else, then OPF (7) can be
written as a standard form QCQP

or included in BFM by modifying (2a) into C

C. OPF as QCQP
Before we describe convex relaxations of OPF, we first show
that when is quadratic in for some where in (10).
Hermitian matrix , OPF is indeed a QCQP by converting it
into the standard form. We will use the derivation in [62] for IV. FEASIBLE SETS AND RELAXATIONS: BIM
OPF (7) in BIM. OPF (9) in BFM can similarly be converted into
a standard form QCQP. In this and the next section, we derive semidefinite relaxations
Define the admittance matrix by of OPF and clarify their relationships. The cost function of
OPF is usually assumed to be convex in its variables. The
difficulty of OPF formulated here thus arises from the nonconvex
feasible sets V for BIM and X for BFM. The basic approach to
deriving convex relaxations of OPF is to design convex supersets
of (equivalent sets of) V or X and minimize the same cost
Here, is symmetric but not necessarily Hermitian. Let be the function over these supersets. Different choices of convex super-
net injection current from bus to the rest of the network. Then sets lead to different relaxations, but they all provide a lower
the current vector and the voltage vector are related by the bound to OPF. If every optimal solution of a convex relaxation
Ohm’s law . BIM (1) is equivalent to happens to lie in V or X then it is also feasible and hence optimal
for the original OPF. In this case, we say the realxation is exact.
In this section, we present three characterizations of the
feasible set V in BIM. These characterizations naturally suggest
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 19

convex supersets and semidefinite relaxations of OPF, and we psd (rank-1), denoted by (rank if,
prove equivalence relationships among them. In the next section, for all edges , the principal submatrices
we treat BFM. In Part II of the paper, we discuss sufficient
conditions that guaranteed exact relaxations.

A. Preliminaries
are psd (rank-1). is a chordal graph if either has no cycle or
Since OPF is a nonconvex QCQP, there is a standard semi-
all its minimal cycles (ones without chords) are of length three. A
definite relaxation through the equivalence relation: for any
chordal extension of is a chordal graph that contains ,
Hermitian matrix , for a psd
i.e., has the same vertex set as but an edge set that is a
rank-1 matrix . Applying this transformation to the QCQP
superset of ’s edge set. In that case, we call the partial matrix
formulation (10) leads to an equivalent problem of the form
a chordal extension of the partial matrix . Every graph
has a chordal extension, generally nonunique. In particular, a
S complete supergraph of is a trivial chordal extension of .
For our purposes, chordal graphs are important because of the
for appropriate Hermitian matrices and real numbers . This result [63, Theorem 7] that every psd partial matrix has a psd
problem is equivalent to (10) because given a psd rank-1 solution completion if and only if the underlying graph is chordal. When a
, a unique solution of (10) can be recovered through rank-1 positive definite completion exists, there is a unique positive
factorization . Unlike (10), which is quadratic in , definite completion, in the class of all positive definite comple-
this problem is convex in except the nonconvex rank-1 tions, whose determinant is maximal. Theorem 2 extends this to
constraint. Removing the rank-1 constraint yields the standard rank-1 partial matrices.
SDP relaxation.
We now generalize this intuition to characterize the feasible B. Feasible Sets
set V in (6) in terms of partial matrices. These characterizations We can now characterize the feasible set V of OPF defined in
lead naturally to SDP, chordal, and SOCP relaxations of OPF in (6). Recall the undirected connected graph that
BIM, as shown in [58] and [28]. models a power network. Given a voltage vector V define a
We start with some basic definitions on partial matrices and partial matrix , for and
their completions; see, e.g., [63]–[65] for more details. Fix any
connected undirected graph with vertices and edges
connecting distinct vertices.1 A partial matrix is a set of
complex numbers defined on

Then, the constraints (5) and (3) imply that the partial matrix
satisfies2

can be interpreted as a matrix with entries partially specified

by these complex numbers. If is a complete graph (in which
there is an edge between every pair of vertices), then is a
fully specified matrix. A completion of is
any fully specified matrix that agrees with on graph
Following Section III-C, these constraints can also be written
, i.e.,
in a (partial) matrix form as

Given an matrix , we use to denote the submatrix

of on , i.e., the partial matrix consisting of the entries of
defined on graph . If is a clique (a fully connected subgraph)
of , then let denote the fully specified principal sub-
matrix of defined on . We extend the definitions of The converse is not always true: given a partial matrix that
Hermitian, psd, and rank-1 for matrices to partial matrices, as satisfies (12), it is not always possible to recover a voltage vector
follows. A partial matrix is Hermitian, denoted by in V. Indeed, this is possible if and only if has a completion
, if for all ; it is psd, that is psd rank-1, because in that case satisfies (12) since
denoted by , if is Hermitian and the principal if and it can be uniquely factored as
submatrices are psd for all cliques of ; it is rank-1, with V. We hence seek conditions additional
denoted by rank , if the principal submatrices to (12) on the partial matrix that guarantee that it has a psd
are rank-1 for all cliques of . We say is
1 2
In this section, we abuse notation and use to denote general integers Constraint (12a) can also be written compactly in terms of admittance matrix
unrelated to the number of buses or lines in a power network. as in [66]
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
20 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

rank-1 completion from which V can be recovered. Our Theorem 2 implies that given a partial matrix W
first key result provides such a characterization. or a partial matrix W there is a psd rank-1 completion
We say that a partial matrix satisfies the cycle condition if W from which a solution V of OPF can be recovered.
for every cycle in In fact we know more, given any Hermitian partial matrix
(not necessarily in W ), the set of all completions of that
satisfies the condition in Theorem 2(3) consists of a single psd
rank-1 matrix and infinitely many indefinite non-rank-1 matri-
ces; see [58, Theorems 5 and 8] and discussions therein. Hence,
When represents voltage phase differences across each
the psd rank-1 completion of a W is unique.
line, then the cycle condition imposes that they sum to zero (mod
) around any cycle. Theorem 2, proved in [58, Theorem 3] and Corollary 3: Given a partial matrix W or
[28], implies that has a psd rank-1 completion if and only W , there is a unique psd rank-1 completion W.
if is psd rank-1 on and satisfies the cycle condition Recall that two sets and are equivalent ( ) if there is
(13), if and only if it has a chordal extension that is psd a bijection between them. Even though W W W are
rank-1.3 different kinds of spaces Theorem 2 and Corollary 3 imply that
Consider the following conditions on they are all equivalent to the feasible set of OPF.
matrices and partial matrices and :
Theorem 4:
V W W W

Theorem 4 suggests three equivalent problems to OPF. We

assume that the cost function in OPF depends on only
Theorem 2: Fix a graph on nodes and any chordal through the partial matrix defined in (9). For example, if the
extension of . Assuming > , > , and cost is total real line loss in the network then
> , , we have .
1) Given an matrix that satisfies (14), If the cost is a weighted sum of real generation power
its submatrix satisfies (15). then where are the given real
2) Given a partial matrix that satisfies (15), its sub- power demands at buses ; again is a function of the
matrix satisfies (16) and the cycle condition (13). partial matrix . Then Theorem 4 implies that OPF (7) is
3) Given a partial matrix that satisfies (16) and the cycle equivalent to
condition (13), there is a completion of that satisfies
(14). W
Informally, Theorem 2 says that (14) is equivalent to (15)
which is equivalent to (16) (13). It characterizes a property of
the full matrix (rank ) in terms of its submatrices where W is any one of the sets W W W . Specifically, given
and . This is important because the submatrices are typically an optimal solution in W, it can be uniquely decomposed
much smaller than for large sparse networks and much easier into . Then is in V and an optimal
to compute. The theorem thus allows us to solve simpler solution of OPF (7). Alternatively given an optimal solution
problems in terms of partial matrices as we now explain. in W or W , Corollary 3 guarantees that has a
Define the set of Hermitian matrices unique psd rank-1 completion in W from which an optimal
V can be recovered. In fact, given a partial matrix
W S W (or W ), there is a more direct
construction of a feasible solution V of OPF than
Fix any chordal extension of and define the set of through its completion; see Section IV-D.
Hermitian partial matrices
C. Semidefinite Relaxations
W
Hence solving OPF (7) is equivalent to solving (20) over any
Finally, define the set of Hermitian partial matrices of W W W for an appropriate matrix variable. The diffi-
culty with solving (20) is that the feasible sets W, W , and W
W are still nonconvex due to the rank-1 constraints and the cycle
condition (13). Their removal leads to SDP, chordal, and SOCP
Note that the definition of psd for partial matrices implies that relaxations of OPF.
and are Hermitian. The assumption > Relax W, W , and W to the following convex supersets:
implies that all matrices or partial matrices have strictly positive
diagonal entries. W S
W
3 W
The theorem also holds with psd replaced by negative semidefinite.
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 21

Define the problems: E. Tightness of Relaxations

OPF-sdp:
Since W W , W W , and W W , the relaxa-
W tions OPF-sdp, OPF-ch, and OPF-socp all provide lower bounds
on OPF (7) in light of Theorem 4. How do these relaxations
OPF-ch: compare in terms of computational efficiency and tightness?
OPF-socp is the simplest computationally. OPF-ch usually
W requires more computation than OPF-socp but much less than
OPF-sdp for large sparse networks (even though OPF-ch can be
OPF-socp: as complex as OPF-sdp in the worse case [64], [65]). The relative
tightness of the relaxations depends on the network topology. For
W a general mesh network, OPF-sdp is as tight a relaxation as
OPF-ch and they are strictly tighter than OPF-socp. For a tree
The condition in the definition of W is (radial) network, the hierarchy collapses and all three are equally
equivalent to and (recall the assumption tight. We now make this precise.
> ) Consider their feasible sets W , W , and W . We say that a
set is an effective subset of a set , denoted by , if, given
> > a (partial) matrix , there is a (partial) matrix that has
the same cost . We say is similar to , denoted by
This is a second-order cone and hence OPF-socp is indeed an , if and . Note that implies
SOCP in the rotated form. but the converse may not be true. The feasible set of OPF (9) is an
Remark 6 (Literature): SOCP relaxation for OPF seems to be effective subset of the feasible sets of the relaxations; moreover,
first proposed in [22] for BIM (1), and in [51] and [57] for BFM these relaxations have similar feasible sets when the network is
(2) as explained in the next section. By defining a new set of radial. This is a slightly different formulation of the same results
variables , , and in [28] and [58].
where , [22] rewrites the BIM Theorem 5: V W W W . If is a tree, then
(1) in the complex domain as a set of linear equations in these V W W W .
new variables in the real domain and the following quadratic
Let be the optimal values of OPF (7),
equations:
OPF-sdp (21), OPF-ch (22), OPF-socp (23), respectively.
Theorems 4 and 5 directly imply the following.
Relaxing these equalities to enlarges the Corollary 6: . If is a tree, then
solution set to a second-order cone that is equivalent to W in this .
paper. SDP relaxation is first proposed in [23] for the BIM and
Remark 7(Tightness): Theorem 5 and Corollary 6 imply that
analyzed in [24]. Chordal relaxation for OPF is first proposed in
for radial networks one should always solve OPF-socp since it is
[25] and [26] and analyzed in [27] and [28].
the tightest and the simplest relaxation of the three. For mesh
networks, there is a tradeoff between OPF-socp and OPF-ch/
D. Solution Recovery
OPF-sdp: the latter is tighter but requires heavier computation.
When the convex relaxations OPF-sdp, OPF-ch, and OPF- Between OPF-ch and OPF-sdp, OPF-ch is usually preferable as
socp are exact, i.e., if their optimal solutions , , and they are equally tight but OPF-ch is usually much faster to solve
happen to lie in W, W , and W , respectively, then an for large sparse networks. See [25]–[28] and [67] for numerical
optimal solution of the original OPF can be recovered from studies that compare these relaxations.
these solutions. Indeed, the recovery method works not just for
F. Chordal Relaxation
an optimal solution, but any feasible solution that lies in W,
W , or W . Moreover, given a W or a W , Theorem 2 through Corollary 6 apply to any chordal extension
the construction of depends on or only through their of . The choice of does not affect the optimal value
submatrix . We hence describe the method for recovering the of the chordal relaxation but determines its complexity. Unfor-
unique from a , which may be a partial matrix in W or the tunately, the optimal choice that minimizes the complexity of
submatrix of a (partial) matrix in W or W . OPF-ch is NP-hard to compute.
Let be an arbitrary spanning tree of rooted at bus 0. Let P This difficulty is due to two conflicting factors in choosing a
denote the unique path from node 0 to node in . Recall that . Recall that the constraint in the definition of
without loss of generality. For , let W consists of multiple constraints that the principal subma-
trices , one for each (maximal) clique of .
When two cliques and share a node, their submatrices
and share entries that must be decoupled
P
by introducing auxiliary variables and equality constraints on
these variables. The choice of determines the number and
Then it can be checked that is in (6) and feasible for OPF. sizes of these submatrices as well as the numbers of
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
22 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

auxiliary variables and additional decoupling constraints. On the define a mapping C R by: for any
one hand, if contains few cliques then the submatrices , with and
tend to be large and expensive to compute [e.g., if .
is the complete graph then there is a single clique, but Throughout this paper, we assume the cost function in
and OPF-ch is identical to OPF-sdp]. On the other OPF (11) depends on only through . For example for
hand, if contains many small cliques then there tends to be total real line loss, . If the cost is a
more overlap and chordal relaxation tends to require more weighted sum of real generation power, then
decoupling constraints. Hence choosing a good chordal exten- where are the real parts of and are the
sion of is important but nontrivial. See [64], [65], and given real power demands at buses ; again depends only
references therein for methods to compute efficient chordal on .
relaxations of general QCQP. For OPF, [27] proposes effective Then the model (24) is a relaxation of BFM (2) in the sense that
techniques to reduce the number of cliques in its chordal the feasible set X of OPF in (9) is an effective subset of X ,
relaxation. To further reduce the problem size, [67] proposes to X X , since X X . We now characterize the subset of
carefully drop some of the decoupling constraints, though the X that is equivalent to X.
resulting relaxation can be weaker. Given an R define R by

V. FEASIBLE SETS AND RELAXATIONS: BFM

We now present an SOCP relaxation of OPF in BFM proposed Even though does not include phase angles of , implies a
in [51] and [57] in two steps. We first relax the phase angles of phase difference across each line given by .
and in (2) and then we relax a set of quadratic equalities to The subset of X that is equivalent to X are those for which
inequalities. This derivation pinpoints the difference between there exists such that . To state this precisely,
radial and mesh topologies. It motivates a recursive version of let be the (transposed) reduced incidence matrix of
BFM for radial networks (Section VI) and the use of phase
shifters for convexification of mesh networks (Part II [30]).

A. Feasible Sets
Consider the following set of equations in the variables where . Consider the set of X such that
in R 4
:

i.e., is in the range space of (mod ). A solution , if

exists, is unique in . Define the set

X R

The following result characterizes the feasible set X of OPF

and define the solution set as
in BFM and follows from [57, Theorems 2, 4].
X R Theorem 7:

Note that the vector includes and includes . The model X X X

(24) is first proposed in [45] and [46].5 It can be derived as a
relaxation of BFM (2) as follows. Taking the squared magnitude The bijection between X and X is given by defined above
of (2c) and replacing and by and , respectively, restricted to X. Its inverse is
yields (24c). To obtain (24b), use (2b)–(2c) to write defined on X in terms of by
and take the squared magnitude on both
sides to eliminate the phase angles of and . These operations

4
The use of complex variables is only a shorthand and should be interpreted as
operations in real variables. For instance, (24a) is a shorthand for The condition (26) is equivalent to the cycle condition (13) in
the BIM. To see this, fix any spanning tree of the
(directed) graph . We can assume without loss of generality
(possibly after re-labeling the links) that consists of links
. Then can be partitioned into

5
The original model, called the DistFlow equations, in [45] and [46] is for radial
(distribution) networks, but its extension here to mesh networks is trivial.
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 23

where the submatrix corresponds to links in and the

submatrix corresponds to links in .
Similarly, partition into

The next result, proved in [57, Theorems 2 and 4], provides a

more explicit characterization of (26) in terms of . When it
holds, this characterization has the same interpretation of the
cycle condition in (13): the voltage angle differences implied by Fig. 1. Feasible sets X of OPF (9) in BFM, its equivalent set X (defined by ) and
sum to zero (mod ) around any cycle. Formally, let be the its relaxations X and X . If is a tree, then X X .
extension of from directed to undirected links: for each
let and . We
say is an undirected cycle if, for each and current angles can be ignored and OPF (11) is equivalent
, either or with to OPF-nc.
the interpretation that ; denotes one Theorem 9: X X X X . If is a tree, then
of these links. X X X X .
Theorem 8: An X satisfies (26) if and only if around Let be the optimal cost of OPF (9) in the BFM. Let ,
each undirected cycle we have , and be the optimal costs of OPF (30), OPF-nc (31),
and OPF-socp (32), respectively, defined above. Theorem 9
implies the following.
Corollary 10: . If is a tree, then
.
In that case, is the unique solution of
(26) in , where projects to . Remark 8 (SOCP Relaxation): Suppose one solves OPF-socp
Theorem 8 determines when the voltage magnitudes of a and obtains an optimal solution X . For
given can be assigned phase angles so that the resulting radial networks, if attains equality in (29) then X
is a power flow solution in X. and Theorem 9 implies that an optimal solution
X of OPF (9) can be recovered from .
B. SOCP Relaxation Indeed, where is defined in (27).
Alternatively, one can use the angle recovery algorithms in
The set X that contains the (equivalent) feasible set X of OPF
[57, Part I] to recover . For mesh networks, needs to
is still nonconvex because of the quadratic equalities in (24c).
both attain equality in (29) and satisfy the cycle condition (26) in
Relax them to inequalities
order for an optimal solution to be recoverable. Our
experience with various practical test networks suggests that
usually attains equality in (29) but, for mesh networks,
and define the set rarely satisfes (26) [28], [51], [56], [57]. Hence, OPF-socp is
effective for radial networks but not for mesh networks (in both
X R BIM and BFM).

Clearly X X X X (see Fig. 1). Moreover, X is a C. Equivalence

second-order cone in the rotated form. Theorem 9 establishes a bijection between X and the feasible
The three sets X, X , and X define the following problems: set X of OPF (9) in BFM. Theorem 4 establishes a bijection
OPF: between W and the feasible set V of OPF (9) in BIM. Theorem 1
hence implies that X X V W . Moreover, their SOCP
X relaxations are equivalent in these two models [28], [58]. Define
the set of partial matrices defined on that are psd rank-1
OPF-nc:
but do not satisfy the cycle condition (13)
X
W
OPF-socp:

X Clearly W W W in general and W W W for

radial networks.
Theorem 9 follows from the results in [57] and implies that Theorem 11:
OPF (11) is equivalent to minimization over X and OPF-socp is
its SOCP relaxation. Moreover for radial networks, voltage X W X W X W
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
24 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

The bijection between X and W is defined as follows. Let Its recursive structure is exploited in [47] to analyze the power
W C denote the set of Hermitian partial matrices flow solutions given an , as we now explain using the
(including which is given). Let special case of a linear network with buses that represents a
denote vectors in R . Define the linear mapping main feeder. To simplify notation, denote and
W by where by and , respectively. Then the DistFlow
equations (33) reduce to ( is given)

Its inverse X W is where Let , . If were known, then one can

for and for . start with and use the recursion (34a)–(34c) to compute
in terms of , i.e., (34) can be collapsed into functions of
The mapping and its inverse restricted to W W and
the scalar variable [recall that are given]
X X defines the bijection between them.

VI. BFM FOR RADIAL NETWORKS

Theorem 9 implies that for radial networks the model (24) is Use the boundary condition (34d) to solve
exact. This is because the reduced incident matrix in (26) is for the scalar variable . The other variables can then be
and invertible, so the cycle condition is always satisfied computed from (35). This method can be extended to a general
[57, Theorem 4]. Hence, a solution in X can be mapped to a radial network with laterals [47]. See also [68] and [69] for
branch flow solution in X by the mapping defined in (27). techniques for solving the nonlinear equations (35), and [54] and
For radial networks, this model has two advantages: 1) it has a [55] for a different recursive approach called the forward/back-
recursive structure that simplifies computation; and 2) it has a ward sweep for radial networks.
linear approximation that provides simple bounds on branch Case II) Links point toward bus 0. Model (24) reduces to
powers and voltage magnitudes , as we now show.

A. Recursive Equations and Graph Orientation

The model (24) holds for any graph orientation of . It has a
recursive structure when is a tree. In that case, different
orientations have different boundary conditions that initialize
the recursion and may be convenient for different applications.
Without loss of generality, we take bus 0 as the root of the where in (36a) denotes the node on the unique path between
tree. We discuss two different orientations: 1) where every node 0 and node . The boundary condition is defined by
link points away from bus 0; and 2) where every link points in (36a) when and in (36a) when is a leaf
toward bus 0.6 node. An advantage of this orientation is illustrated in Sec-
tion VI-B in proving a simple bound on .
Case I) Links point away from bus 0. Model (24) reduces to
B. Linear Approximation and Bounds
By setting in (33), we obtain a linear approximation of
the BFM, with the graph orientation where all links point away
from bus 0:

where bus in (33a) denotes the unique parent of node (on the
unique path from node 0 to node ), with the understanding that if
then and . Similarly, when is a leaf node7
all in (33a). The model (33) is called the DistFlow where bus in (37a) denotes the unique parent of bus . The
equations and first proposed in [45] and [46]. boundary condition is in (37a) when , and
6 in (37a) when is a leaf node. This is called the
An alternative model is to use an undirected graph and, for each link , the
variables and are defined for both directions, with the simplified DistFlow equations in [46] and [70]. It is a good
additional equations and . approximation of (33) because the loss is typically much
7
A node is a leaf node if there exists no such that . smaller than the branch power flow .
Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 25

The next result provides simple bounds on in terms of

their linear approximations . Denote by T the subtree
rooted at bus , including . We write “ T ” to mean node of
T and “ T ” to mean edge of T . Denote by P the
set of links on the unique path from bus 0 to bus .
Lemma 12: Fix any and R . Let and
be solutions of (33) and (37), respectively, with the
given and . Then
1) For

T Fig. 2. Solving OPF through semidefinite relaxations.

T T 2) For all , with equality if and only if all

in T are zero.
2) For , with equality if only if and all 3) For
in T are zero.
3) For
P

P P

P
4) For , .
Lemma 13 says that the branch power (toward bus 0)
equals the total injection T in the subtree rooted at bus
4) For , .
Lemma 12 says that the power flow on line equals the minus the line loss in that subtree. The linear approximation
total load in the subtree rooted at node plus the total neglects the loss and hence overestimates the branch power flow.
T
line loss in supplying these loads. The linear approximation It can be proved by recursing on (36a)–(36b) and (38).
neglects the line loss and underestimates the required power to Remark 9: Bounds for SOCP relaxation. Lemmas 12 and 13 do
supply these loads. not depend on the quadratic equalities (33c) and (36c) as long as
Lemma 12(1)–(3) can be easily proved by recursing on . In particular, the lemmas hold if the equalities have been
(33a)–(33b) and (37). Since but , a direct relaxed to inequalities . These bounds are used in
proof of Lemma 12(4) is not obvious. Instead, one can make use [56] to prove a sufficient condition for exact SOCP relaxation for
of Lemma 13 and define a bijection between the solutions radial networks.
of (33) and the solutions of (36) in which
. It can be checked that the solutions of (37) and those of Remark 10 (Linear Approximations): For radial networks,
(38) are related by and . Then Lemma 13 the linear approximations (37) and (38) of BFM have two
(4) implies Lemma 12(4). advantages over the (linear) DC approximation of BIM. First,
A linear approximation of (36) is (setting ) they have a simple recursive structure that leads to simple bounds
on power flow quantities. Second, DC approximation assumes a
lossless network ( ), fixes voltage magnitudes, and ignores
reactive power, whereas (37) and (38) do not. This is important
for distribution systems where loss is much higher than in
transmission systems, voltages can fluctuate significantly, and
reactive powers are used to regulate them. On the other hand, (37)
and (38) are applicable only for radial networks whereas DC
Lemma 13: Fix any and R . Let and approximation applies to mesh networks as well. See also [21]
be solutions of (36) and (38) respectively with the for a more accurate linearization of BIM that addresses the
given and . Then shortcomings of DC OPF.
1) For all

VII. CONCLUSION
T
We have presented a BIM and a BFM, formulated several
relaxations of OPF, and proved their relationships. These results
T T suggest a new approach to solving OPF summarized in Fig. 2.

Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
26 TRANSACTIONS ON CONTROL OF NETWORK SYSTEMS

For radial networks, we recommend solving OPF-socp in [21] C. Coffrin and P. Van Hentenryck, “A linear-programming approximation
either BIM or BFM though there is preliminary evidence that of AC power flows,” CoRR, abs/1206.3614, 2012.
[22] R. A. Jabr, “Radial distribution load flow using conic programming,” IEEE
BFM can be more stable numerically. For mesh networks, we Trans. Power Syst., vol. 21, no. 3, pp. 1458–1459, Aug. 2006.
recommend solving OPF-ch for small networks and OPF-socp [23] X. Bai, H. Wei, K. Fujisawa, and Y. Wang, “Semidefinite programming for
followed by a heuristic search for a feasible point for large optimal power flow problems,” Int. J. Elect. Power Energy Syst., vol. 30,
no. 6–7, pp. 383–392, 2008.
networks. [24] J. Lavaei and S. H. Low, “Zero duality gap in optimal power flow problem,”
The key for this solution strategy is that the relaxations are IEEE Trans. Power Syst., vol. 27, no. 1, pp. 92–107, Feb. 2012.
exact so that an optimal solution of the original OPF can be [25] X. Bai and H. Wei, “A semidefinite programming method with graph
partitioning technique for optimal power flow problems,” Int. J. Elect.
recovered. In Part II of this paper [30], we summarize sufficient Power Energy Syst., vol. 33, no. 7, pp. 1309–1314, 2011.
conditions that guarantee exact relaxation. [26] R. A. Jabr, “Exploiting sparsity in SDP relaxations of the OPF problem,”
IEEE Trans. Power Syst., vol. 27, no. 2, pp. 1138–1139, May 2012.
[27] D. Molzahn, J. Holzer, B. Lesieutre, and C. DeMarco, “Implementation of a
ACKNOWLEDGMENT large-scale optimal power flow solver based on semidefinite programming,”
IEEE Trans. Power Syst., vol. 28, no. 4, pp. 3987–3998, Nov. 2013.
We thank the anonymous reviewers for their helpful [28] S. Bose, S. H. Low, M. Chandy, T. Teeraratkul, and B. Hassibi, “Equivalent
suggestions. relaxations of optimal power flow,” IEEE Trans. Automatic Control, 2014,
to be published.
[29] Z. Luo, W. Ma, A. M. C. So, Y. Ye, and S. Zhang, “Semidefinite relaxation
REFERENCES of quadratic optimization problems,” IEEE Signal Process. Mag., vol. 27,
no. 3, pp. 20–34, May 2010.
[1] S. H. Low, “Convex relaxation of optimal power flow: A tutorial,” in Proc. [30] S. H. Low, “Convex relaxation of optimal power flow, II: Exactness,” IEEE
IREP Symp. Bulk Power Syst. Dyn. Control (IREP), Rethymnon, Greece, Trans. Control Netw. Syst., 2014, to be published.
Aug. 2013. [31] P. P. Varaiya, F. F. Wu, and J. W. Bialek, “Smart operation of smart grid:
[2] J. Carpentier, “Contribution to the economic dispatch problem,” Bull. Soc. Risk-limiting dispatch,” Proc. IEEE, vol. 99, no. 1, pp. 40–57, Jan. 2011.
Francoise Electr., vol. 3, no. 8, pp. 431–447, 1962 (in French). [32] M. Vrakopoulou, M. Katsampani, K. Margellos, J. Lygeros, and
[3] H. W. Dommel and W. F. Tinney, “Optimal power flow solutions,” IEEE G. Andersson, “Probabilistic security-constrained ac optimal power flow,”
Trans. Power App. Syst., vol. 87, no. 10, pp. 1866–1876, Oct. 1968. in Proc. IEEE Grenoble PowerTech (POWERTECH), Jun. 2013, pp. 1–6.
[4] J. A. Momoh, Electric Power System Applications of Optimization. Power [33] R. Bent, D. Bienstock, and M. Chertkov, “Synchronization-aware and
Engineering. New York, NY, USA: Markel Dekker, 2001. algorithm-efficient chance constrained optimal power flow,” in Proc. IREP
[5] M. Huneault and F. D. Galiana, “A survey of the optimal power flow Symp. Bulk Power Syst. Dyn. Control (IREP), Rethymnon, Greece,
literature,” IEEE Trans. Power Syst., vol. 6, no. 2, pp. 762–770, May 1991. Aug. 2013.
[6] J. A. Momoh, M. E. El-Hawary, and R. Adapa, “A review of selected [34] B. H. Kim and R. Baldick, “Coarse-grained distributed optimal power
optimal power flow literature to 1993. Part I: Nonlinear and quadratic flow,” IEEE Trans. Power Syst., vol. 12, no. 2, pp. 932–939, May 1997.
programming approaches,” IEEE Trans. Power Syst., vol. 14, no. 1, [35] R. Baldick, B. H. Kim, C. Chase, and Y. Luo, “A fast distributed imple-
pp. 96–104, Feb. 1999. mentation of optimal power flow,” IEEE Trans. Power Syst., vol. 14, no. 3,
[7] J. A. Momoh, M. E. El-Hawary, and R. Adapa, “A review of selected pp. 858–864, Aug. 1999.
optimal power flow literature to 1993. Part II: Newton, linear programming [36] G. Hug-Glanzmann and G. Andersson, “Decentralized optimal power flow
and interior point methods,” IEEE Trans. Power Syst., vol. 14, no. 1, control for overlapping areas in power systems,” IEEE Trans. Power Syst.,
pp. 105–111, Feb. 1999. vol. 24, no. 1, pp. 327–336, Feb. 2009.
[8] K. S. Pandya and S. K. Joshi, “A survey of optimal power flow methods,” [37] F. J. Nogales, F. J. Prieto, and A. J. Conejo, “A decomposition methodology
J. Theor. Appl. Inf. Technol., vol. 4, no. 5, pp. 450–458, 2008. applied to the multi-area optimal power flow problem,” Ann. Oper. Res.,
[9] S. Frank, I. Steponavice, and S. Rebennack, “Optimal power flow: vol. 120, pp. 99–116, 2003.
A bibliographic survey, I: Formulations and deterministic methods,” Ener- [38] A. Lam, B. Zhang, and D. N. Tse, “Distributed algorithms for optimal
gy Syst., vol. 3, pp. 221–258, Sep. 2012. power flow problem,” in Proc. IEEE 51st Annu. Conf. Decis. Control
[10] S. Frank, I. Steponavice, and S. Rebennack, “Optimal power flow: (CDC), 2012, pp. 430–437.
A bibliographic survey, II: Nondeterministic and hybrid methods,” Energy [39] M. Kraning, E. Chu, J. Lavaei, and S. Boyd, “Dynamic network energy
Syst., vol. 3, pp. 259–289, Sep. 2013. management via proximal message passing,” Found. Trends Optim., vol. 1,
[11] M. B. Cain, R. P. O’Neill, and A. Castillo, “History of optimal power flow no. 2, pp. 70–122, 2013.
and formulations (OPF Paper 1),” Tech. Rep., US FERC, Dec. 2012. [40] K. Turitsyn, P. Sǔlc, S. Backhaus, and M. Chertkov, “Options for control of
[12] R. P. O’Neill, A. Castillo, and M. B. Cain, “The IV formulation and linear reactive power by distributed photovoltaic generators,” Proc. IEEE, vol. 99,
approximations of the AC optimal power flow problem (OPF Paper 2),” no. 6, pp. 1063–1073, Jun. 2011.
Tech. Rep., US FERC, Dec. 2012. [41] A. Y. S. Lam, B. Zhang, A. Domínguez-García, and D. Tse, “Optimal
[13] R. P. O’Neill, A. Castillo, and M. B. Cain, “The computational testing of AC distributed voltage regulation in power distribution networks,” arXiv,
optimal power flow using the current voltage formulations (OPF Paper 3),” Apr. 2012.
Tech. Rep., US FERC, Dec. 2012. [42] D. T. Phan, “Lagrangian duality and branch-and-bound algorithms
[14] A. Castillo and R. P. O’Neill, “Survey of approaches to solving the ACOPF for optimal power flow,” Oper. Res., vol. 60, no. 2, pp. 275–285,
(OPF Paper 4),” Tech. Rep., US FERC, Mar. 2013. Mar./Apr. 2012.
[15] A. Castillo and R. P. O’Neill, “Computational performance of solution [43] A. Gopalakrishnan, A. U. Raghunathan, D. Nikovski, and L. T. Biegler,
techniques applied to the ACOPF (OPF Paper 5),” Tech. Rep., US FERC, “Global optimization of optimal power flow using a branch amp; bound
Mar. 2013. algorithm,” in Proc. 50th Annu. Allerton Conf. Commun. Control Comput.,
[16] B. Stott and O. Alsaç, “Fast decoupled load flow,” IEEE Trans. Power App. Oct. 2012, pp. 609–616.
Syst., vol. 93, no. 3, pp. 859–869, May 1974. [44] C. Josz, J. Maeght, P. Panciatici, and J. C. Gilbert, “Application of the
[17] O. Alsaç, J. Bright, M. Prais, and B. Stott, “Further developments in moment? SOS approach to global optimization of the OPF problem,” arXiv,
LP-based optimal power flow,” IEEE Trans. Power Syst., vol. 5, no. 3, Nov. 2013.
pp. 697–711, Aug. 1990. [45] M. E. Baran and F. F. Wu, “Optimal capacitor placement on radial
[18] T. J. Overbye, X. Cheng, and Y. Sun, “A comparison of the AC, and DC distribution systems,” IEEE Trans. Power Del., vol. 4, no. 1, pp. 725–734,
power flow models for LMP calculations,” in Proc. 37th Hawaii Int. Conf. Nov. 1989.
Syst. Sci., 2004, pp. 5–8. [46] M. E. Baran and F. F. Wu, “Optimal sizing of capacitors placed on a radial
[19] K. Purchala, L. Meeus, D. Van Dommelen, and R. Belmans, “Usefulness of distribution system,” IEEE Trans. Power Del., vol. 4, no. 1, pp. 735–743,
DC power flow for active power flow analysis,” in Proc. IEEE PES General Jan. 1989.
Meeting, 2005, pp. 2457–2462. [47] H.-D. Chiang and M. E. Baran, “On the existence and uniqueness of load
[20] B. Stott, J. Jardim, and O. Alsaç, “DC power flow revisited,” IEEE Trans. flow solution for radial distribution power networks,” IEEE Trans. Circuits
Power Syst., vol. 24, no. 3, pp. 1290–1300, Aug. 2009. Syst., vol. 37, no. 3, pp. 410–416, Mar. 1990.

Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.
LOW: CONVEX RELAXATION OF OPF 27

[48] H.-D. Chiang, “A decoupled load flow method for distribution power [64] M. Fukuda, M. Kojima, K. Murota, and K. Nakata, “Exploiting sparsity in
networks: Algorithms, analysis and convergence study,” Int. J. Elect. Power semidefinite programming via matrix completion I: General framework,”
Energy Syst., vol. 13, no. 3, pp. 130–138, Jun. 1991. SIAM J. Optim., vol. 11, pp. 647–674, 2001.
[49] R. Cespedes, “New method for the analysis of distribution networks,” IEEE [65] K. Nakata, K. Fujisawa, M. Fukuda, M. Kojima, and K. Murota, “Exploiting
Trans. Power Del., vol. 5, no. 1, pp. 391–396, Jan. 1990. sparsity in semidefinite programming via matrix completion II: Implemen-
[50] A. G. Expósito and E. R. Ramos, “Reliable load flow technique for tation and numerical results,” Math. Program., vol. 95, no. 2, pp. 303–327,
radial distribution networks,” IEEE Trans. Power Syst., vol. 14, no. 13, 2003.
pp. 1063–1069, Aug. 1999. [66] B. Zhang and D. Tse, “Geometry of the injection region of power networks,”
[51] M. Farivar, C. R. Clarke, S. H. Low, and K. Mani Chandy, “Inverter IEEE Trans. Power Syst., vol. 28, no. 2, pp. 788–797, May 2013.
VAR control for distribution systems with renewables,” in Proc. IEEE [67] M. S. Andersen, A. Hansson, and L. Vandenberghe, “Reduced-
SmartGridComm Conf., Oct. 2011, pp. 457–462. complexity semidefinite relaxations of optimal power flow problems, ”
[52] J. A. Taylor and F. S. Hover, “Convex models of distribution system arXiv:1308.6718v1, Aug. 2013.
reconfiguration,” IEEE Trans. Power Syst., vol. 27, no. 3, pp. 1407–1413, [68] R. D. Zimmerman and H.-D. Chiang, “Fast decoupled power flow for
Aug. 2012. unbalanced radial distribution systems,” IEEE Trans. Power Syst., vol. 10,
[53] J. Lavaei, D. Tse, and B. Zhang, “Geometry of power flows and optimiza- no. 4, pp. 2045–2052, Nov. 1995.
tion in distribution networks,” IEEE Trans. Power Syst., vol. 29, no. 2, [69] M. S. Srinivas, “Distribution load flows: A brief review,” in Proc. IEEE
pp. 572–583, 2014. Power Eng. Soc. Winter Meeting, 2000, vol. 2, pp. 942–945.
[54] W. H. Kersting, Distribution Systems Modeling and Analysis. Boca Raton, [70] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution systems
FL, USA: CRC Press, 2002. for loss reduction and load balancing,” IEEE Trans. Power Del., vol. 4,
[55] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, “A no. 2, pp. 1401–1407, Apr. 1989.
compensation-based power flow method for weakly meshed distribution
and transmission networks,” IEEE Trans. Power Syst., vol. 3, no. 2,
pp. 753–762, May 1988. Steven H. Low (F’08) received the B.S. degree from
[56] L. Gan, N. Li, U. Topcu, and S. H. Low, “Exact convex relaxation of optimal Cornell University, Ithaca, NY, USA, and the Ph.D.
power flow in radial networks,” IEEE Trans. Automat. Control, 2014. degree from the University of California, Berkeley,
[57] M. Farivar and S. H. Low, “Branch flow model: Relaxations and convex- CA, USA, both in electrical engineering.
ification (parts I, II),” IEEE Trans. Power Syst., vol. 28, no. 3, He is a Professor of Computing + Mathematical
pp. 2554–2572, Aug. 2013. Sciences and Electrical Engineering, Caltech,
[58] S. Bose, S. H. Low, and M. Chandy, “Equivalence of branch flow, and bus Pasadena, CA, USA, and a Changjiang Chair
injection models,” in Proc. 50th Annu. Allerton Conf. Commun. Control Professor with Zhejiang University, Hangzhou,
Comput., Oct. 2012, pp. 1893–1899. China. Before that, he was with AT&T Bell Labs,
[59] D. Gan, R. J. Thomas, and R. D. Zimmerman, “Stability-constrained Murray Hill, NJ, USA, and the University of Mel-
optimal power flow,” IEEE Trans. Power Syst., vol. 15, no. 2, pp. 535–540, bourne, Victoria, Australia. His research interests
May 2000. include power systems and communication networks.
[60] F. Capitanescua, J. L. Martinez Ramosb, P. Panciaticic, D. Kirschend, Dr. Low was a co-recipient of IEEE best paper awards, the R&D 100 Award,
A. Marano Marcolinie, L. Platbroodf, and L. Wehenkelg, “State-of-the-art, and an Okawa Foundation Research Grant. He is on the Technical Advisory
challenges, and future trends in security constrained optimal power flow,” Board of Southern California Edison and was a member of the Networking and
Electr. Power Syst. Res., vol. 81, no. 8, pp. 1731–1741, Aug. 2011. Information Technology Technical Advisory Group for the US President’s
[61] B. Stott and O. Alsaç, “Optimal power flow: Basic requirements for real-life Council of Advisors on Science and Technology (PCAST). He is a Senior Editor
problems and their solutions,” White paper, Jul. 2012. of the IEEE Journal on Selected Areas in Communications, the IEEE TRANSAC-
[62] S. Bose, D. Gayme, K. M. Chandy, and S. H. Low, “Quadratically TIONS ON CONTROL OF NETWORK SYSTEMS, and the IEEE TRANSACTIONS ON NETWORK
constrained quadratic programs on acyclic graphs with application to SCIENCE AND ENGINEERING and is on the editorial board of NOW Foundations and
power flow,” arXiv:1203.5599v1, Mar. 2012. Trends in Networking and in Electric Energy Systems.
[63] R. Grone, C. R. Johnson, E. M. Sá, and H. Wolkowicz, “Positive definite
completions of partial Hermitian matrices,” Linear Algebr. Appl., vol. 58,
pp. 109–124, 1984.

Authorized licensed use limited to: National Chung Cheng University. Downloaded on October 24,2024 at 16:22:03 UTC from IEEE Xplore. Restrictions apply.