PDF version (NAG web site
, 64bit version, 64bit version)
NAG Toolbox: nag_opt_qpconvex1_sparse_solve (e04nk)
Purpose
nag_opt_qpconvex1_sparse_solve (e04nk) solves sparse linear programming or quadratic programming problems.
Syntax
[
ns,
xs,
istate,
miniz,
minz,
ninf,
sinf,
obj,
clamda,
user,
lwsav,
iwsav,
rwsav,
ifail] = e04nk(
n,
m,
iobj,
ncolh,
qphx,
a,
ha,
ka,
bl,
bu,
start,
names,
crname,
ns,
xs,
istate,
leniz,
lenz,
lwsav,
iwsav,
rwsav, 'nnz',
nnz, 'nname',
nname, 'user',
user)
[
ns,
xs,
istate,
miniz,
minz,
ninf,
sinf,
obj,
clamda,
user,
lwsav,
iwsav,
rwsav,
ifail] = nag_opt_qpconvex1_sparse_solve(
n,
m,
iobj,
ncolh,
qphx,
a,
ha,
ka,
bl,
bu,
start,
names,
crname,
ns,
xs,
istate,
leniz,
lenz,
lwsav,
iwsav,
rwsav, 'nnz',
nnz, 'nname',
nname, 'user',
user)
Before calling
nag_opt_qpconvex1_sparse_solve (e04nk), or
the option setting function
(e04nm),
nag_opt_init (e04wb) must be called.
Description
nag_opt_qpconvex1_sparse_solve (e04nk) is designed to solve a class of quadratic programming problems that are assumed to be stated in the following general form:
where
x$x$ is a set of variables,
A$A$ is an
m$m$ by
n$n$ matrix and the objective function
f(x)$f\left(x\right)$ may be specified in a variety of ways depending upon the particular problem to be solved. The optional parameter
Maximize may be used to specify an alternative problem in which
f(x)$f\left(x\right)$ is maximized. The possible forms for
f(x)$f\left(x\right)$ are listed in
Table 1, in which the prefixes FP, LP and QP stand for ‘feasible point’, ‘linear programming’ and ‘quadratic programming’ respectively,
c$c$ is an
n$n$element vector and
H$H$ is the
n$n$ by
n$n$ secondderivative matrix
∇^{2}f(x)${\nabla}^{2}f\left(x\right)$ (the
Hessian matrix).
Problem type 
Objective function f(x)$f\left(x\right)$ 
Hessian matrix H$H$ 
FP 
Not applicable 
Not applicable 
LP 
c^{T}x${c}^{\mathrm{T}}x$ 
Not applicable 
QP 
c^{T}x + (1/2)x^{T}Hx${c}^{\mathrm{T}}x+\frac{1}{2}{x}^{\mathrm{T}}Hx$ 
Symmetric positive semidefinite 
Table 1
For LP and QP problems, the unique global minimum value of
f(x)$f\left(x\right)$ is found. For FP problems,
f(x)$f\left(x\right)$ is omitted and the function attempts to find a feasible point for the set of constraints. For QP problems, you must also provide a function that computes
Hx$Hx$ for any given vector
x$x$. (
H$H$ need not be stored explicitly.) If
H$H$ is the zero matrix, the function will still solve the resulting LP problem; however, this can be accomplished more efficiently by setting
ncolh = 0${\mathbf{ncolh}}=0$ (see
Section [Parameters]).
The defining feature of a
convex QP problem is that the matrix
H$H$ must be
positive semidefinite, i.e., it must satisfy
x^{T}Hx ≥ 0${x}^{\mathrm{T}}Hx\ge 0$ for all
x$x$. Otherwise,
f(x)$f\left(x\right)$ is said to be
nonconvex and it may be more appropriate to call
nag_opt_nlp1_sparse_solve (e04ug) instead.
nag_opt_qpconvex1_sparse_solve (e04nk) is intended to solve largescale linear and quadratic programming problems in which the constraint matrix
A$A$ is
sparse (i.e., when the number of zero elements is sufficiently large that it is worthwhile using algorithms which avoid computations and storage involving zero elements). The function also takes advantage of sparsity in
c$c$. (Sparsity in
H$H$ can be exploited in the function that computes
Hx$Hx$.) For problems in which
A$A$ can be treated as a
dense matrix, it is usually more efficient to use
nag_opt_lp_solve (e04mf),
nag_opt_lsq_lincon_solve (e04nc) or
nag_opt_qp_dense_solve (e04nf).
The upper and lower bounds on the
m$m$ elements of
Ax$Ax$ are said to define the
general constraints of the problem. Internally,
nag_opt_qpconvex1_sparse_solve (e04nk) converts the general constraints to equalities by introducing a set of
slack variables s$s$, where
s = (s_{1},s_{2}, … ,s_{m})^{T}$s={({s}_{1},{s}_{2},\dots ,{s}_{m})}^{\mathrm{T}}$. For example, the linear constraint
5 ≤ 2x_{1} + 3x_{2} ≤ + ∞$5\le 2{x}_{1}+3{x}_{2}\le +\infty $ is replaced by
2x_{1} + 3x_{2} − s_{1} = 0$2{x}_{1}+3{x}_{2}{s}_{1}=0$, together with the bounded slack
5 ≤ s_{1} ≤ + ∞$5\le {s}_{1}\le +\infty $. The problem defined by
(1) can therefore be rewritten in the following equivalent form:
Since the slack variables
s$s$ are subject to the same upper and lower bounds as the elements of
Ax$Ax$, the bounds on
Ax$Ax$ and
x$x$ can simply be thought of as bounds on the combined vector
(x,s)$(x,s)$. (In order to indicate their special role in QP problems, the original variables
x$x$ are sometimes known as ‘column variables’, and the slack variables
s$s$ are known as ‘row variables’.)
Each LP or QP problem is solved using an activeset method. This is an iterative procedure with two phases: a feasibility phase, in which the sum of infeasibilities is minimized to find a feasible point; and an optimality phase, in which f(x)$f\left(x\right)$ is minimized by constructing a sequence of iterations that lies within the feasible region.
A constraint is said to be active or binding at x$x$ if the associated element of either x$x$ or Ax$Ax$ is equal to one of its upper or lower bounds. Since an active constraint in Ax$Ax$ has its associated slack variable at a bound, the status of both simple and general upper and lower bounds can be conveniently described in terms of the status of the variables (x,s)$(x,s)$. A variable is said to be nonbasic if it is temporarily fixed at its upper or lower bound. It follows that regarding a general constraint as being active is equivalent to thinking of its associated slack as being nonbasic.
At each iteration of an activeset method, the constraints
Ax − s = 0$Axs=0$ are (conceptually) partitioned into the form
where
x_{N}${x}_{N}$ consists of the nonbasic elements of
(x,s)$(x,s)$ and the
basis matrix B$B$ is square and nonsingular. The elements of
x_{B}${x}_{B}$ and
x_{S}${x}_{S}$ are called the
basic and
superbasic variables respectively; with
x_{N}${x}_{N}$ they are a permutation of the elements of
x$x$ and
s$s$. At a QP solution, the basic and superbasic variables will lie somewhere between their upper or lower bounds, while the nonbasic variables will be equal to one of their bounds. At each iteration,
x_{S}${x}_{S}$ is regarded as a set of independent variables that are free to move in any desired direction, namely one that will improve the value of the objective function (or sum of infeasibilities). The basic variables are then adjusted in order to ensure that
(x,s)$(x,s)$ continues to satisfy
Ax − s = 0$Axs=0$. The number of superbasic variables (
n_{S}${n}_{S}$ say) therefore indicates the number of degrees of freedom remaining after the constraints have been satisfied. In broad terms,
n_{S}${n}_{S}$ is a measure of
how nonlinear the problem is. In particular,
n_{S}${n}_{S}$ will always be zero for FP and LP problems.
If it appears that no improvement can be made with the current definition of B$B$, S$S$ and N$N$, a nonbasic variable is selected to be added to S$S$, and the process is repeated with the value of n_{S}${n}_{S}$ increased by one. At all stages, if a basic or superbasic variable encounters one of its bounds, the variable is made nonbasic and the value of n_{S}${n}_{S}$ is decreased by one.
Associated with each of the
m$m$ equality constraints
Ax − s = 0$Axs=0$ is a
dual variable π_{i}${\pi}_{i}$. Similarly, each variable in
(x,s)$(x,s)$ has an associated
reduced gradient d_{j}${d}_{j}$ (also known as a
reduced cost). The reduced gradients for the variables
x$x$ are the quantities
g − A^{T}π$g{A}^{\mathrm{T}}\pi $, where
g$g$ is the gradient of the QP objective function; and the reduced gradients for the slack variables
s$s$ are the dual variables
π$\pi $. The QP subproblem is optimal if
d_{j} ≥ 0${d}_{j}\ge 0$ for all nonbasic variables at their lower bounds,
d_{j} ≤ 0${d}_{j}\le 0$ for all nonbasic variables at their upper bounds and
d_{j} = 0${d}_{j}=0$ for all superbasic variables. In practice, an
approximate QP solution is found by slightly relaxing these conditions on
d_{j}${d}_{j}$ (see the description of the optional parameter
Optimality Tolerance).
The process of computing and comparing reduced gradients is known as
pricing (a term first introduced in the context of the simplex method for linear programming). To ‘price’ a nonbasic variable
x_{j}${x}_{j}$ means that the reduced gradient
d_{j}${d}_{j}$ associated with the relevant active upper or lower bound on
x_{j}${x}_{j}$ is computed via the formula
d_{j} = g_{j} − a^{T}π${d}_{j}={g}_{j}{a}^{\mathrm{T}}\pi $, where
a_{j}${a}_{j}$ is the
j$j$th column of
$\left(\begin{array}{cc}A& I\end{array}\right)$. (The variable selected by such a process and the corresponding value of
d_{j}${d}_{j}$ (i.e., its reduced gradient) are the quantities
+S and
dj in the monitoring file output; see
Section [Description of Monitoring Information].) If
A$A$ has significantly more columns than rows (i.e.,
n ≫ m$n\gg m$), pricing can be computationally expensive. In this case, a strategy known as
partial pricing can be used to compute and compare only a subset of the
d_{j}${d}_{j}$'s.
nag_opt_qpconvex1_sparse_solve (e04nk) is based on SQOPT, which is part of the SNOPT package described in
Gill et al. (2002), which in turn utilizes functions from the MINOS package (see
Murtagh and Saunders (1995)). It uses stable numerical methods throughout and includes a reliable basis package (for maintaining sparse
LU$LU$ factors of the basis matrix
B$B$), a practical antidegeneracy procedure, efficient handling of linear constraints and bounds on the variables (by an activeset strategy), as well as automatic scaling of the constraints. Further details can be found in
Section [Algorithmic Details].
References
Fourer R (1982) Solving staircase linear programs by the simplex method Math. Programming 23 274–313
Gill P E and Murray W (1978) Numerically stable methods for quadratic programming Math. Programming 14 349–372
Gill P E, Murray W and Saunders M A (2002) SNOPT: An SQP Algorithm for Largescale Constrained Optimization 12 979–1006 SIAM J. Optim.
Gill P E, Murray W, Saunders M A and Wright M H (1987) Maintaining LU factors of a general sparse matrix Linear Algebra and its Applics. 88/89 239–270
Gill P E, Murray W, Saunders M A and Wright M H (1989) A practical anticycling procedure for linearly constrained optimization Math. Programming 45 437–474
Gill P E, Murray W, Saunders M A and Wright M H (1991) Inertiacontrolling methods for general quadratic programming SIAM Rev. 33 1–36
Hall J A J and McKinnon K I M (1996) The simplest examples where the simplex method cycles and conditions where EXPAND fails to prevent cycling Report MS 96–100 Department of Mathematics and Statistics, University of Edinburgh
Murtagh B A and Saunders M A (1995) MINOS 5.4 users' guide Report SOL 8320R Department of Operations Research, Stanford University
Parameters
Compulsory Input Parameters
 1:
n – int64int32nag_int scalar
n$n$, the number of variables (excluding slacks). This is the number of columns in the linear constraint matrix A$A$.
Constraint:
n ≥ 1${\mathbf{n}}\ge 1$.
 2:
m – int64int32nag_int scalar
m$m$, the number of general linear constraints (or slacks). This is the number of rows in
A$A$, including the free row (if any; see
iobj).
Constraint:
m ≥ 1${\mathbf{m}}\ge 1$.
 3:
iobj – int64int32nag_int scalar
If
iobj > 0${\mathbf{iobj}}>0$, row
iobj of
A$A$ is a free row containing the nonzero elements of the vector
c$c$ appearing in the linear objective term
c^{T}x${c}^{\mathrm{T}}x$.
If
iobj = 0${\mathbf{iobj}}=0$, there is no free row, i.e., the problem is either an FP problem (in which case
iobj must be set to zero), or a QP problem with
c = 0$c=0$.
Constraint:
0 ≤ iobj ≤ m$0\le {\mathbf{iobj}}\le {\mathbf{m}}$.
 4:
ncolh – int64int32nag_int scalar
n_{H}${n}_{H}$, the number of leading nonzero columns of the Hessian matrix
H$H$. For FP and LP problems,
ncolh must be set to zero.
Constraint:
0 ≤ ncolh ≤ n$0\le {\mathbf{ncolh}}\le {\mathbf{n}}$.
 5:
qphx – function handle or string containing name of mfile
For QP problems, you must supply a version of
qphx to compute the matrix product
Hx$Hx$. If
H$H$ has zero rows and columns, it is most efficient to order the variables
x = (yz) $x={\left(\begin{array}{cc}y& z\end{array}\right)}^{\mathrm{T}}$ so that
where the nonlinear variables
y$y$ appear first as shown. For FP and LP problems,
qphx will never be called by
nag_opt_qpconvex1_sparse_solve (e04nk) and hence
qphx may be the string
'e54nku'.
[hx, user] = qphx(nstate, ncolh, x, user)
Input Parameters
 1:
nstate – int64int32nag_int scalar
If
nstate = 1${\mathbf{nstate}}=1$,
nag_opt_qpconvex1_sparse_solve (e04nk) is calling
qphx for the first time. This parameter setting allows you to save computation time if certain data must be read or calculated only once.
If
nstate ≥ 2${\mathbf{nstate}}\ge 2$,
nag_opt_qpconvex1_sparse_solve (e04nk) is calling
qphx for the last time. This parameter setting allows you to perform some additional computation on the final solution. In general, the last call to
qphx is made with
nstate = 2 + ifail${\mathbf{nstate}}=2+{\mathbf{ifail}}$ (see
Section [Error Indicators and Warnings]).
Otherwise,
nstate = 0${\mathbf{nstate}}=0$.
 2:
ncolh – int64int32nag_int scalar
This is the same parameter
ncolh as supplied to
nag_opt_qpconvex1_sparse_solve (e04nk).
 3:
x(ncolh) – double array
The first
ncolh elements of the vector
x$x$.
 4:
user – Any MATLAB object
qphx is called from
nag_opt_qpconvex1_sparse_solve (e04nk) with the object supplied to
nag_opt_qpconvex1_sparse_solve (e04nk).
Output Parameters
 1:
hx(ncolh) – double array
The product Hx$Hx$.
 2:
user – Any MATLAB object
 6:
a(nnz) – double array
nnz, the dimension of the array, must satisfy the constraint
1 ≤ nnz ≤ n × m$1\le {\mathbf{nnz}}\le {\mathbf{n}}\times {\mathbf{m}}$.
The nonzero elements of A$A$, ordered by increasing column index. Note that elements with the same row and column indices are not allowed.
 7:
ha(nnz) – int64int32nag_int array
nnz, the dimension of the array, must satisfy the constraint
1 ≤ nnz ≤ n × m$1\le {\mathbf{nnz}}\le {\mathbf{n}}\times {\mathbf{m}}$.
ha(i)${\mathbf{ha}}\left(\mathit{i}\right)$ must contain the row index of the nonzero element stored in
a(i)${\mathbf{a}}\left(\mathit{i}\right)$, for
i = 1,2, … ,nnz$\mathit{i}=1,2,\dots ,{\mathbf{nnz}}$. Note that the row indices for a column may be supplied in any order.
Constraint:
1 ≤ ha(i) ≤ m$1\le {\mathbf{ha}}\left(\mathit{i}\right)\le {\mathbf{m}}$, for
i = 1,2, … ,nnz$\mathit{i}=1,2,\dots ,{\mathbf{nnz}}$.
 8:
ka(n + 1${\mathbf{n}}+1$) – int64int32nag_int array
ka(j)${\mathbf{ka}}\left(\mathit{j}\right)$ must contain the index in
a of the start of the
j$\mathit{j}$th column, for
j = 1,2, … ,n$\mathit{j}=1,2,\dots ,{\mathbf{n}}$.
ka(n + 1)${\mathbf{ka}}\left({\mathbf{n}}+1\right)$ must be set to
nnz + 1${\mathbf{nnz}}+1$. To specify the
j$j$th column as empty, set
ka(j) = ka(j + 1)${\mathbf{ka}}\left(j\right)={\mathbf{ka}}\left(j+1\right)$. As a consequence
ka(1)${\mathbf{ka}}\left(1\right)$ is always
1$1$.
Constraints:
 ka(1) = 1${\mathbf{ka}}\left(1\right)=1$;
 ka(j) ≥ 1${\mathbf{ka}}\left(\mathit{j}\right)\ge 1$, for j = 2,3, … ,n$\mathit{j}=2,3,\dots ,{\mathbf{n}}$;
 ka(n + 1) = nnz + 1${\mathbf{ka}}\left({\mathbf{n}}+1\right)={\mathbf{nnz}}+1$;
 0 ≤ ka(j + 1) − ka(j) ≤ m$0\le {\mathbf{ka}}\left(\mathit{j}+1\right){\mathbf{ka}}\left(\mathit{j}\right)\le {\mathbf{m}}$, for j = 1,2, … ,n$\mathit{j}=1,2,\dots ,{\mathbf{n}}$.
 9:
bl(n + m${\mathbf{n}}+{\mathbf{m}}$) – double array
l$l$, the lower bounds for all the variables and general constraints, in the following order. The first
n elements of
bl must contain the bounds on the variables
x$x$, and the next
m elements the bounds for the general linear constraints
Ax$Ax$ (or slacks
s$s$) and the free row (if any). To specify a nonexistent lower bound (i.e.,
l_{j} = − ∞${l}_{j}=\infty $), set
bl(j) ≤ − bigbnd${\mathbf{bl}}\left(j\right)\le \mathit{bigbnd}$, where
bigbnd$\mathit{bigbnd}$ is the value of the optional parameter
Infinite Bound Size. To specify the
j$j$th constraint as an
equality, set
bl(j) = bu(j) = β${\mathbf{bl}}\left(j\right)={\mathbf{bu}}\left(j\right)=\beta $, say, where
β < bigbnd$\left\beta \right<\mathit{bigbnd}$. Note that the lower bound corresponding to the free row must be set to
− ∞$\infty $ and stored in
bl(n + iobj)${\mathbf{bl}}\left({\mathbf{n}}+{\mathbf{iobj}}\right)$.
Constraint:
if
iobj > 0${\mathbf{iobj}}>0$,
bl(n + iobj) ≤ − bigbnd${\mathbf{bl}}\left({\mathbf{n}}+{\mathbf{iobj}}\right)\le \mathit{bigbnd}$(See also the description for
bu.)
 10:
bu(n + m${\mathbf{n}}+{\mathbf{m}}$) – double array
u$u$, the upper bounds for all the variables and general constraints, in the following order. The first
n elements of
bu must contain the bounds on the variables
x$x$, and the next
m elements the bounds for the general linear constraints
Ax$Ax$ (or slacks
s$s$) and the free row (if any). To specify a nonexistent upper bound (i.e.,
u_{j} = + ∞${u}_{j}=+\infty $), set
bu(j) ≥ bigbnd${\mathbf{bu}}\left(j\right)\ge \mathit{bigbnd}$. Note that the upper bound corresponding to the free row must be set to
+ ∞$+\infty $ and stored in
bu(n + iobj)${\mathbf{bu}}\left({\mathbf{n}}+{\mathbf{iobj}}\right)$.
Constraints:
 if iobj > 0${\mathbf{iobj}}>0$, bu(n + iobj) ≥ bigbnd${\mathbf{bu}}\left({\mathbf{n}}+{\mathbf{iobj}}\right)\ge \mathit{bigbnd}$;
 bl(j) ≤ bu(j)${\mathbf{bl}}\left(\mathit{j}\right)\le {\mathbf{bu}}\left(\mathit{j}\right)$, for j = 1,2, … ,n + m$\mathit{j}=1,2,\dots ,{\mathbf{n}}+{\mathbf{m}}$;
 if bl(j) = bu(j) = β${\mathbf{bl}}\left(j\right)={\mathbf{bu}}\left(j\right)=\beta $, β < bigbnd$\left\beta \right<\mathit{bigbnd}$.
 11:
start – string (length ≥ 1)
Indicates how a starting basis is to be obtained.
 start = 'C'${\mathbf{start}}=\text{'C'}$
 An internal Crash procedure will be used to choose an initial basis matrix B$B$.
 start = 'W'${\mathbf{start}}=\text{'W'}$
 A basis is already defined in istate (probably from a previous call).
Constraint:
start = 'C'${\mathbf{start}}=\text{'C'}$ or
'W'$\text{'W'}$.
 12:
names(5$5$) – cell array of strings
A set of names associated with the socalled MPSX form of the problem, as follows:
 names(1)${\mathbf{names}}\left(1\right)$
 Must contain the name for the problem (or be blank).
 names(2)${\mathbf{names}}\left(2\right)$
 Must contain the name for the free row (or be blank).
 names(3)${\mathbf{names}}\left(3\right)$
 Must contain the name for the constraint righthand side (or be blank).
 names(4)${\mathbf{names}}\left(4\right)$
 Must contain the name for the ranges (or be blank).
 names(5)${\mathbf{names}}\left(5\right)$
 Must contain the name for the bounds (or be blank).
(These names are used in the monitoring file output; see
Section [Description of Monitoring Information].)
 13:
crname(nname) – cell array of strings
nname, the dimension of the array, must satisfy the constraint
nname = 1${\mathbf{nname}}=1$ or
n + m${\mathbf{n}}+{\mathbf{m}}$.
The optional column and row names, respectively.
If
nname = 1${\mathbf{nname}}=1$,
crname is not referenced and the printed output will use default names for the columns and rows.
If
nname = n + m${\mathbf{nname}}={\mathbf{n}}+{\mathbf{m}}$, the first
n elements must contain the names for the columns and the next
m elements must contain the names for the rows. Note that the name for the free row (if any) must be stored in
crname(n + iobj)${\mathbf{crname}}\left({\mathbf{n}}+{\mathbf{iobj}}\right)$.
 14:
ns – int64int32nag_int scalar
n_{S}${n}_{S}$, the number of superbasics. For QP problems,
ns need not be specified if
start = 'C'${\mathbf{start}}=\text{'C'}$, but must retain its value from a previous call when
start = 'W'${\mathbf{start}}=\text{'W'}$. For FP and LP problems,
ns need not be initialized.
 15:
xs(n + m${\mathbf{n}}+{\mathbf{m}}$) – double array
The initial values of the variables and slacks
(x,s)$(x,s)$. (See the description for
istate.)
 16:
istate(n + m${\mathbf{n}}+{\mathbf{m}}$) – int64int32nag_int array
If
start = 'C'${\mathbf{start}}=\text{'C'}$, the first
n elements of
istate and
xs must specify the initial states and values, respectively, of the variables
x$x$. (The slacks
s$s$ need not be initialized.) An internal Crash procedure is then used to select an initial basis matrix
B$B$. The initial basis matrix will be triangular (neglecting certain small elements in each column). It is chosen from various rows and columns of
$\left(\begin{array}{cc}A& I\end{array}\right)$. Possible values for
istate(j)${\mathbf{istate}}\left(j\right)$ are as follows:
istate(j)${\mathbf{istate}}\left(j\right)$  State of xs(j)${\mathbf{xs}}\left(j\right)$ during Crash procedure 
0$0$ or 1$1$  Eligible for the basis 
2$2$  Ignored 
3$3$  Eligible for the basis (given preference over 0$0$ or 1$1$) 
4$4$ or 5$5$  Ignored 
If nothing special is known about the problem, or there is no wish to provide special information, you may set
istate(j) = 0${\mathbf{istate}}\left(\mathit{j}\right)=0$ and
xs(j) = 0.0${\mathbf{xs}}\left(\mathit{j}\right)=0.0$, for
j = 1,2, … ,n$\mathit{j}=1,2,\dots ,{\mathbf{n}}$. All variables will then be eligible for the initial basis. Less trivially, to say that the
j$\mathit{j}$th variable will probably be equal to one of its bounds, set
istate(j) = 4${\mathbf{istate}}\left(j\right)=4$ and
xs(j) = bl(j)${\mathbf{xs}}\left(j\right)={\mathbf{bl}}\left(j\right)$ or
istate(j) = 5${\mathbf{istate}}\left(j\right)=5$ and
xs(j) = bu(j)${\mathbf{xs}}\left(j\right)={\mathbf{bu}}\left(j\right)$ as appropriate.
Following the Crash procedure, variables for which
istate(j) = 2${\mathbf{istate}}\left(j\right)=2$ are made superbasic. Other variables not selected for the basis are then made nonbasic at the value
xs(j)${\mathbf{xs}}\left(j\right)$ if
bl(j) ≤ xs(j) ≤ bu(j)${\mathbf{bl}}\left(j\right)\le {\mathbf{xs}}\left(j\right)\le {\mathbf{bu}}\left(j\right)$, or at the value
bl(j)${\mathbf{bl}}\left(j\right)$ or
bu(j)${\mathbf{bu}}\left(j\right)$ closest to
xs(j)${\mathbf{xs}}\left(j\right)$.
If
start = 'W'${\mathbf{start}}=\text{'W'}$,
istate and
xs must specify the initial states and values, respectively, of the variables and slacks
(x,s)$(x,s)$. If
nag_opt_qpconvex1_sparse_solve (e04nk) has been called previously with the same values of
n and
m,
istate already contains satisfactory information.
Constraints:
 if start = 'C'${\mathbf{start}}=\text{'C'}$, 0 ≤ istate(j) ≤ 5$0\le {\mathbf{istate}}\left(\mathit{j}\right)\le 5$, for j = 1,2, … ,n$\mathit{j}=1,2,\dots ,{\mathbf{n}}$;
 if start = 'W'${\mathbf{start}}=\text{'W'}$, 0 ≤ istate(j) ≤ 3$0\le {\mathbf{istate}}\left(\mathit{j}\right)\le 3$, for j = 1,2, … ,n + m$\mathit{j}=1,2,\dots ,{\mathbf{n}}+{\mathbf{m}}$.
 17:
leniz – int64int32nag_int scalar
The dimension of the array
iz as declared in the (sub)program from which
nag_opt_qpconvex1_sparse_solve (e04nk) is called.
Constraint:
leniz ≥ 1${\mathbf{leniz}}\ge 1$.
 18:
lenz – int64int32nag_int scalar
The dimension of the array
z as declared in the (sub)program from which
nag_opt_qpconvex1_sparse_solve (e04nk) is called.
Constraint:
lenz ≥ 1${\mathbf{lenz}}\ge 1$.
The amounts of workspace provided (i.e.,
leniz and
lenz) and required (i.e.,
miniz and
minz) are (by default for
nag_opt_qpconvex1_sparse_solve (e04nk)) output on the current advisory message unit
nadv (as defined by
nag_file_set_unit_advisory (x04ab)). Since the minimum values of
leniz and
lenz required to start solving the problem are returned in
miniz and
minz, respectively, you may prefer to obtain appropriate values from the output of a preliminary run with
leniz and
lenz set to
1$1$. (
nag_opt_qpconvex1_sparse_solve (e04nk) will then terminate with
ifail = 12${\mathbf{ifail}}={\mathbf{12}}$.)
 19:
lwsav(20$20$) – logical array
 20:
iwsav(380$380$) – int64int32nag_int array
 21:
rwsav(285$285$) – double array
The arrays
lwsav,
iwsav and
rwsav must not be altered between calls to any of the functions
nag_opt_qpconvex1_sparse_solve (e04nk),
(e04nl),
(e04nm) or
nag_opt_init (e04wb).
Optional Input Parameters
 1:
nnz – int64int32nag_int scalar
Default:
The dimension of the arrays
a,
ha. (An error is raised if these dimensions are not equal.)
The number of nonzero elements in A$A$.
Constraint:
1 ≤ nnz ≤ n × m$1\le {\mathbf{nnz}}\le {\mathbf{n}}\times {\mathbf{m}}$.
 2:
nname – int64int32nag_int scalar
Default:
The dimension of the array
crname.
The number of column (i.e., variable) and row names supplied in
crname.
 nname = 1${\mathbf{nname}}=1$
 There are no names. Default names will be used in the printed output.
 nname = n + m${\mathbf{nname}}={\mathbf{n}}+{\mathbf{m}}$
 All names must be supplied.
Constraint:
nname = 1${\mathbf{nname}}=1$ or
n + m${\mathbf{n}}+{\mathbf{m}}$.
 3:
user – Any MATLAB object
user is not used by
nag_opt_qpconvex1_sparse_solve (e04nk), but is passed to
qphx. Note that for large objects it may be more efficient to use a global variable which is accessible from the mfiles than to use
user.
Input Parameters Omitted from the MATLAB Interface
 iz z iuser ruser
Output Parameters
 1:
ns – int64int32nag_int scalar
The final number of superbasics. This will be zero for FP and LP problems.
 2:
xs(n + m${\mathbf{n}}+{\mathbf{m}}$) – double array
The final values of the variables and slacks (x,s)$(x,s)$.
 3:
istate(n + m${\mathbf{n}}+{\mathbf{m}}$) – int64int32nag_int array
The final states of the variables and slacks
(x,s)$(x,s)$. The significance of each possible value of
istate(j)${\mathbf{istate}}\left(j\right)$ is as follows:
istate(j)${\mathbf{istate}}\left(j\right)$  State of variable j$j$  Normal value of xs(j)${\mathbf{xs}}\left(j\right)$ 
0$0$  Nonbasic  bl(j)${\mathbf{bl}}\left(j\right)$ 
1$1$  Nonbasic  bu(j)${\mathbf{bu}}\left(j\right)$ 
2$2$  Superbasic  Between bl(j)${\mathbf{bl}}\left(j\right)$ and bu(j)${\mathbf{bu}}\left(j\right)$ 
3$3$  Basic  Between bl(j)${\mathbf{bl}}\left(j\right)$ and bu(j)${\mathbf{bu}}\left(j\right)$ 
If
ninf = 0${\mathbf{ninf}}=0$, basic and superbasic variables may be outside their bounds by as much as the value of the optional parameter
Feasibility Tolerance. Note that unless the
Scale Option = 0${\mathbf{Scale\; Option}}=0$ is specified, the optional parameter
Feasibility Tolerance applies to the variables of the scaled problem. In this case, the variables of the original problem may be as much as
0.1$0.1$ outside their bounds, but this is unlikely unless the problem is very badly scaled.
Very occasionally some nonbasic variables may be outside their bounds by as much as the optional parameter
Feasibility Tolerance, and there may be some nonbasic variables for which
xs(j)${\mathbf{xs}}\left(j\right)$ lies strictly between its bounds.
If
ninf > 0${\mathbf{ninf}}>0$, some basic and superbasic variables may be outside their bounds by an arbitrary amount (bounded by
sinf if
Scale Option = 0${\mathbf{Scale\; Option}}=0$).
 4:
miniz – int64int32nag_int scalar
The minimum value of
leniz required to start solving the problem. If
ifail = 12${\mathbf{ifail}}={\mathbf{12}}$,
nag_opt_qpconvex1_sparse_solve (e04nk) may be called again with
leniz suitably larger than
miniz. (The bigger the better, since it is not certain how much workspace the basis factors need.)
 5:
minz – int64int32nag_int scalar
The minimum value of
lenz required to start solving the problem. If
ifail = 13${\mathbf{ifail}}={\mathbf{13}}$,
nag_opt_qpconvex1_sparse_solve (e04nk) may be called again with
lenz suitably larger than
minz. (The bigger the better, since it is not certain how much workspace the basis factors need.)
 6:
ninf – int64int32nag_int scalar
The number of infeasibilities. This will be zero if
ifail = 0${\mathbf{ifail}}={\mathbf{0}}$ or
1${\mathbf{1}}$.
 7:
sinf – double scalar
The sum of infeasibilities. This will be zero if
ninf = 0${\mathbf{ninf}}=0$. (Note that
nag_opt_qpconvex1_sparse_solve (e04nk) does
not attempt to compute the minimum value of
sinf if
ifail = 3${\mathbf{ifail}}={\mathbf{3}}$.)
 8:
obj – double scalar
The value of the objective function.
If
ninf = 0${\mathbf{ninf}}=0$,
obj includes the quadratic objective term
(1/2)x^{T}Hx$\frac{1}{2}{x}^{\mathrm{T}}Hx$ (if any).
If
ninf > 0${\mathbf{ninf}}>0$,
obj is just the linear objective term
c^{T}x${c}^{\mathrm{T}}x$ (if any).
For FP problems,
obj is set to zero.
 9:
clamda(n + m${\mathbf{n}}+{\mathbf{m}}$) – double array
A set of Lagrange multipliers for the bounds on the variables and the general constraints. More precisely, the first
n elements contain the multipliers (
reduced costs) for the bounds on the variables, and the next
m elements contain the multipliers (
shadow prices) for the general linear constraints.
 10:
user – Any MATLAB object
 11:
lwsav(20$20$) – logical array
 12:
iwsav(380$380$) – int64int32nag_int array
 13:
rwsav(285$285$) – double array
 14:
ifail – int64int32nag_int scalar
ifail = 0${\mathrm{ifail}}={\mathbf{0}}$ unless the function detects an error (see
[Error Indicators and Warnings]).
nag_opt_qpconvex1_sparse_solve (e04nk) returns with
ifail = 0${\mathbf{ifail}}={\mathbf{0}}$ if the reduced gradient (
Norm rg; see
Section [Printed output]) is negligible, the Lagrange multipliers (
Lagr Mult; see
Section [Printed output]) are optimal and
x$x$ satisfies the constraints to the accuracy requested by the value of the optional parameter
Feasibility Tolerance (
default value = max (10^{ − 6},sqrt(ε))$\text{default value}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}({10}^{6},\sqrt{\epsilon})$, where
ε$\epsilon $ is the
machine precision).
Error Indicators and Warnings
Note: nag_opt_qpconvex1_sparse_solve (e04nk) may return useful information for one or more of the following detected errors or warnings.
Errors or warnings detected by the function:
Cases prefixed with W are classified as warnings and
do not generate an error of type NAG:error_n. See nag_issue_warnings.
 W ifail = 1${\mathbf{ifail}}=1$
Weak solution found. The final x$x$ is not unique, although x$x$ gives the global minimum value of the objective function.
 W ifail = 2${\mathbf{ifail}}=2$
The problem is unbounded (or badly scaled). The objective function is not bounded below in the feasible region.
 W ifail = 3${\mathbf{ifail}}=3$
The problem is infeasible. The general constraints cannot all be satisfied simultaneously to within the value of the optional parameter
Feasibility Tolerance (
default value = max (10^{ − 6},sqrt(ε))$\text{default value}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}({10}^{6},\sqrt{\epsilon})$, where
ε$\epsilon $ is the
machine precision).
 ifail = 4${\mathbf{ifail}}=4$
Too many iterations. The value of the optional parameter
Iteration Limit (
default value = max (50,5(n + m))$\text{default value}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}(50,5(n+m))$) is too small.
 ifail = 5${\mathbf{ifail}}=5$
The reduced Hessian matrix
Z^{T}HZ${Z}^{\mathrm{T}}HZ$ (see
Section [Definition of the Working Set and Search Direction]) exceeds its assigned dimension. The value of the optional parameter
Superbasics Limit (
default value = min (n_{H} + 1,n)$\text{default value}=\mathrm{min}\phantom{\rule{0.125em}{0ex}}({n}_{H}+1,n)$) is too small.
 ifail = 6${\mathbf{ifail}}=6$
The Hessian matrix
H$H$ appears to be indefinite. This sometimes occurs because the values of the optional parameters
LU Factor Tolerance (
default value = 100.0$\text{default value}=100.0$) and
LU Update Tolerance (
default value = 10.0$\text{default value}=10.0$) are too large. Check also that
qphx has been coded correctly and that all relevant elements of
Hx$Hx$ have been assigned their correct values.
 ifail = 7${\mathbf{ifail}}=7$

An input parameter is invalid.
 ifail = 8${\mathbf{ifail}}=8$
Numerical error in trying to satisfy the general constraints. The basis is very illconditioned.
 ifail = 9${\mathbf{ifail}}=9$
Not enough integer workspace for the basis factors. Increase
leniz and rerun
nag_opt_qpconvex1_sparse_solve (e04nk).
 ifail = 10${\mathbf{ifail}}=10$
Not enough real workspace for the basis factors. Increase
lenz and rerun
nag_opt_qpconvex1_sparse_solve (e04nk).
 ifail = 11${\mathbf{ifail}}=11$
The basis is singular after
15$15$ attempts to factorize it (adding slacks where necessary). Either the problem is badly scaled or the value of the optional parameter
LU Factor Tolerance (
default value = 100.0$\text{default value}=100.0$) is too large.
 ifail = 12${\mathbf{ifail}}=12$
Not enough integer workspace to start solving the problem. Increase
leniz to at least
miniz and rerun
nag_opt_qpconvex1_sparse_solve (e04nk).
 ifail = 13${\mathbf{ifail}}=13$
Not enough real workspace to start solving the problem. Increase
lenz to at least
minz and rerun
nag_opt_qpconvex1_sparse_solve (e04nk).
Accuracy
nag_opt_qpconvex1_sparse_solve (e04nk) implements a numerically stable activeset strategy and returns solutions that are as accurate as the condition of the problem warrants on the machine.
Further Comments
This section contains a description of the printed output.
Description of the Printed Output
The following line of summary output (
< 80$\text{}<80$ characters) is produced at every iteration. In all cases, the values of the quantities printed are those in effect
on
completion of the given iteration.
Itn 
is the iteration count.

Step 
is the step taken along the computed search direction.

Ninf 
is the number of violated constraints (infeasibilities). This will be zero during the optimality phase.

Sinf/Objective 
is the value of the current objective function. If x$x$ is not feasible, Sinf gives the sum of the magnitudes of constraint violations. If x$x$ is feasible, Objective is the value of the objective function. The output line for the final iteration of the feasibility phase (i.e., the first iteration for which Ninf is zero) will give the value of the true objective at the first feasible point. During the optimality phase, the value of the objective function will be nonincreasing. During the feasibility phase, the number of constraint infeasibilities will not increase until either a feasible point is found, or the optimality of the multipliers implies that no feasible point exists.

Norm rg 
is ‖d_{S}‖$\Vert {d}_{S}\Vert $, the Euclidean norm of the reduced gradient (see Section [Main Iteration]). During the optimality phase, this norm will be approximately zero after a unit step. For FP and LP problems, Norm rg is not printed.

The final printout includes a listing of the status of every variable and constraint.
The following describes the printout for each variable. A full stop (.) is printed for any numerical value that is zero.
Variable 
gives the name of the variable. If nname = 1${\mathbf{nname}}=1$, a default name is assigned to the j$\mathit{j}$th variable, for j = 1,2, … ,n$\mathit{j}=1,2,\dots ,n$. If nname = n + m${\mathbf{nname}}={\mathbf{n}}+{\mathbf{m}}$, the name supplied in crname(j)${\mathbf{crname}}\left(\mathit{j}\right)$ is assigned to the j$\mathit{j}$th variable.

State 
gives the state of the variable (LL if nonbasic on its lower bound, UL if nonbasic on its upper bound, EQ if nonbasic and fixed, FR if nonbasic and strictly between its bounds, BS if basic and SBS if superbasic).
A key is sometimes printed before State.
Note that unless the optional parameter Scale Option = 0${\mathbf{Scale\; Option}}=0$ ( default value = 2$\text{default value}=2$) is specified, the tests for assigning a key are applied to the variables of the scaled problem.
A 
Alternative optimum possible. The variable is nonbasic, but its reduced gradient is essentially zero. This means that if the variable were allowed to start moving away from its bound, there would be no change in the value of the objective function. The values of the other free variables might change, giving a genuine alternative solution. However, if there are any degenerate variables (labelled D), the actual change might prove to be zero, since one of them could encounter a bound immediately. In either case, the values of the Lagrange multipliers might also change.

D 
Degenerate. The variable is basic or superbasic, but it is equal (or very close) to one of its bounds.

I 
Infeasible. The variable is basic or superbasic and is currently violating one of its bounds by more than the value of the Feasibility Tolerance.

N 
Not precisely optimal. The variable is nonbasic or superbasic. If the value of the reduced gradient for the variable exceeds the value of the optional parameter Optimality Tolerance, the solution would not be declared optimal because the reduced gradient for the variable would not be considered negligible.


Value 
is the value of the variable at the final iteration.

Lower Bound 
is the lower bound specified for the variable. None indicates that bl(j) ≤ − bigbnd${\mathbf{bl}}\left(j\right)\le \mathit{bigbnd}$.

Upper Bound 
is the upper bound specified for the variable. None indicates that bu(j) ≥ bigbnd${\mathbf{bu}}\left(j\right)\ge \mathit{bigbnd}$.

Lagr Mult 
is the Lagrange multiplier for the associated bound. This will be zero if State is FR. If x$x$ is optimal, the multiplier should be nonnegative if State is LL, nonpositive if State is UL and zero if State is BS or SBS.

Residual 
is the difference between the variable Value and the nearer of its (finite) bounds bl(j)${\mathbf{bl}}\left(j\right)$ and bu(j)${\mathbf{bu}}\left(j\right)$. A blank entry indicates that the associated variable is not bounded (i.e., bl(j) ≤ − bigbnd${\mathbf{bl}}\left(j\right)\le \mathit{bigbnd}$ and bu(j) ≥ bigbnd${\mathbf{bu}}\left(j\right)\ge \mathit{bigbnd}$).

The meaning of the printout for linear constraints is the same as that given above for variables, with ‘variable’ replaced by ‘constraint’,
n$n$ replaced by
m$m$,
crname(j)${\mathbf{crname}}\left(j\right)$ replaced by
crname(n + j)${\mathbf{crname}}\left(n+j\right)$,
bl(j)${\mathbf{bl}}\left(j\right)$ and
bu(j)${\mathbf{bu}}\left(j\right)$ are replaced by
bl(n + j)${\mathbf{bl}}\left(n+j\right)$ and
bu(n + j)${\mathbf{bu}}\left(n+j\right)$ respectively, and with the following change in the heading:
Constrnt 
gives the name of the linear constraint.

Note that movement off a constraint (as opposed to a variable moving away from its bound) can be interpreted as allowing the entry in the Residual column to become positive.
Numerical values are output with a fixed number of digits; they are not guaranteed to be accurate to this precision.
Example
Open in the MATLAB editor:
nag_opt_qpconvex1_sparse_solve_example
function nag_opt_qpconvex1_sparse_solve_example
n = int64(7);
m = int64(8);
iobj = int64(8);
ncolh = int64(7);
a = [0.02;
0.02;
0.03;
1;
0.7;
0.02;
0.15;
200;
0.06;
0.75;
0.03;
0.04;
0.05;
0.04;
1;
2000;
0.02;
1;
0.01;
0.08;
0.08;
0.8;
2000;
1;
0.12;
0.02;
0.02;
0.75;
0.04;
2000;
0.01;
0.8;
0.02;
1;
0.02;
0.06;
0.02;
2000;
1;
0.01;
0.01;
0.97;
0.01;
400;
0.97;
0.03;
1;
400];
ha = [int64(7);5;3;1;6;4;2;8;7;6;5;4;3;2;1;8;2;1;4;3;7;6;8;1;7;3;4;6;2;8;5;6;7;1;2;3;4;8;1;2;3;6;7;8;7;2;1;8];
ka = [int64(1);9;17;24;31;39;45;49];
bl = [0;
0;
400;
100;
0;
0;
0;
2000;
1e25;
1e25;
1e25;
1e25;
1500;
250;
1e25];
bu = [200;
2500;
800;
700;
1500;
1e25;
1e25;
2000;
60;
100;
40;
30;
1e25;
300;
1e25];
start = 'C';
names = {' '; ' '; ' '; ' '; ' '};
crname = {'...X1...'; '...X2...'; '...X3...'; '...X4...'; '...X5...'; ...
'...X6...'; '...X7...'; '..ROW1..'; '..ROW2..'; '..ROW3..'; ...
'..ROW4..'; '..ROW5..'; '..ROW6..'; '..ROW7..'; '..COST..'};
ns = int64(1232765364);
xs = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
istate = zeros(15, 1, 'int64');
leniz = int64(10000);
lenz = int64(10000);
[cwsav,lwsav,iwsav,rwsav,ifail] = ...
nag_opt_init('nag_opt_qpconvex1_sparse_solve');
[nsOut, xsOut, istateOut, miniz, minz, ninf, sinf, obj, clamda, user, lwsavOut, iwsavOut, rwsavOut, ifail] = ...
nag_opt_qpconvex1_sparse_solve(n, m, iobj, ncolh, @qphx, a, ha, ka, bl, bu, start, ...
names, crname, ns, xs, istate, leniz, lenz, ...
lwsav, iwsav, rwsav);
obj
function [hx, user] = qphx(nstate, ncolh, x, user)
hx = zeros(ncolh, 1);
hx(1) = 2*x(1);
hx(2) = 2*x(2);
hx(3) = 2*(x(3)+x(4));
hx(4) = hx(3);
hx(5) = 2*x(5);
hx(6) = 2*(x(6)+x(7));
hx(7) = hx(6);
obj =
1.8478e+06
Open in the MATLAB editor:
e04nk_example
function e04nk_example
n = int64(7);
m = int64(8);
iobj = int64(8);
ncolh = int64(7);
a = [0.02;
0.02;
0.03;
1;
0.7;
0.02;
0.15;
200;
0.06;
0.75;
0.03;
0.04;
0.05;
0.04;
1;
2000;
0.02;
1;
0.01;
0.08;
0.08;
0.8;
2000;
1;
0.12;
0.02;
0.02;
0.75;
0.04;
2000;
0.01;
0.8;
0.02;
1;
0.02;
0.06;
0.02;
2000;
1;
0.01;
0.01;
0.97;
0.01;
400;
0.97;
0.03;
1;
400];
ha = [int64(7);5;3;1;6;4;2;8;7;6;5;4;3;2;1;8;2;1;4;3;7;6;8;1;7;3;4;6;2; ...
8;5;6;7;1;2;3;4;8;1;2;3;6;7;8;7;2;1;8];
ka = [int64(1);9;17;24;31;39;45;49];
bl = [0;0;400;100;0;0;0;2000;1e25;1e25;1e25;1e25;1500;250;1e25];
bu = [200;2500;800;700;1500;1e25;1e25;2000;60;100;40;30;1e25;300;1e25];
start = 'C';
names = {' '; ' '; ' '; ' '; ' '};
crname = {'...X1...'; '...X2...'; '...X3...'; '...X4...'; '...X5...'; ...
'...X6...'; '...X7...'; '..ROW1..'; '..ROW2..'; '..ROW3..'; ...
'..ROW4..'; '..ROW5..'; '..ROW6..'; '..ROW7..'; '..COST..'};
ns = int64(1232765364);
xs = [0;0;0;0;0;0;0;0;0;0;0;0;0;0;0];
istate = zeros(15, 1, 'int64');
leniz = int64(10000);
lenz = int64(10000);
[cwsav,lwsav,iwsav,rwsav,ifail] = e04wb('e04nk');
[nsOut, xsOut, istateOut, miniz, minz, ninf, sinf, obj, clamda, user, ...
lwsavOut, iwsavOut, rwsavOut, ifail] = ...
e04nk(n, m, iobj, ncolh, @qphx, a, ha, ka, bl, bu, start, ...
names, crname, ns, xs, istate, leniz, lenz, ...
lwsav, iwsav, rwsav);
obj
function [hx, user] = qphx(nstate, ncolh, x, user)
hx = zeros(ncolh, 1);
hx(1) = 2*x(1);
hx(2) = 2*x(2);
hx(3) = 2*(x(3)+x(4));
hx(4) = hx(3);
hx(5) = 2*x(5);
hx(6) = 2*(x(6)+x(7));
hx(7) = hx(6);
obj =
1.8478e+06
Note: the remainder of this document is intended for more advanced users. Section [Algorithmic Details] contains a detailed description of the algorithm which may be needed in order to understand Sections [Optional Parameters] and [Description of Monitoring Information]. Section [Optional Parameters] describes the optional parameters which may be set by calls to nag_opt_qpconvex1_sparse_option_string (e04nm). Section [Description of Monitoring Information] describes the quantities which can be requested to monitor the course of the computation.
Algorithmic Details
This section contains a detailed description of the method used by nag_opt_qpconvex1_sparse_solve (e04nk).
Overview
nag_opt_qpconvex1_sparse_solve (e04nk) is based on an inertiacontrolling method that maintains a Cholesky factorization of the reduced Hessian (see below). The method is similar to that of
Gill and Murray (1978), and is described in detail by
Gill et al. (1991). Here we briefly summarise the main features of the method. Where possible, explicit reference is made to the names of variables that are parameters of the function or appear in the printed output.
The method used has two distinct phases: finding an initial feasible point by minimizing the sum of infeasibilities (the
feasibility phase), and minimizing the quadratic objective function within the feasible region (the
optimality phase). The computations in both phases are performed by the same functions. The twophase nature of the algorithm is reflected by changing the function being minimized from the sum of infeasibilities (the printed quantity
Sinf; see
Section [Description of Monitoring Information]) to the quadratic objective function (the printed quantity
Objective; see
Section [Description of Monitoring Information]).
In general, an iterative process is required to solve a quadratic program. Given an iterate
(x,s)$(x,s)$ in both the original variables
x$x$ and the slack variables
s$s$, a new iterate
(x,s)$(\stackrel{}{x},\stackrel{}{s})$ is defined by
where the
step length
α$\alpha $ is a nonnegative scalar (the printed quantity
Step; see
Section [Description of Monitoring Information]), and
p$p$ is called the
search direction. (For simplicity, we shall consider a typical iteration and avoid reference to the index of the iteration.) Once an iterate is feasible (i.e., satisfies the constraints), all subsequent iterates remain feasible.
Definition of the Working Set and Search Direction
At each iterate
(x,s)$(x,s)$, a
working set of constraints is defined to be a linearly independent subset of the constraints that are satisfied ‘exactly’ (to within the value of the optional parameter
Feasibility Tolerance). The working set is the current prediction of the constraints that hold with equality at a solution of the LP or QP problem. Let
m_{W}${m}_{W}$ denote the number of constraints in the working set (including bounds), and let
W$W$ denote the associated
m_{W}${m}_{W}$ by
(n + m)$(n+m)$ working set matrix consisting of the
m_{W}${m}_{W}$ gradients of the working set constraints.
The search direction is defined so that constraints in the working set remain
unaltered for any value of the step length. It follows that
p$p$ must satisfy the identity
This characterisation allows
p$p$ to be computed using any
n$n$ by
n_{Z}${n}_{Z}$ fullrank matrix
Z$Z$ that spans the null space of
W$W$. (Thus,
n_{Z} = n − m_{W}${n}_{Z}=n{m}_{W}$ and
WZ = 0$WZ=0$.) The null space matrix
Z$Z$ is defined from a sparse
LU$LU$ factorization of part of
W$W$ (see
(6) and
(7)). The direction
p$p$ will satisfy
(3) if
where
p_{Z}${p}_{Z}$ is any
n_{Z}${n}_{Z}$vector.
The working set contains the constraints
Ax − s = 0$Axs=0$ and a subset of the upper and lower bounds on the variables
(x,s)$(x,s)$. Since the gradient of a bound constraint
x_{j} ≥ l_{j}${x}_{j}\ge {l}_{j}$ or
x_{j} ≤ u_{j}${x}_{j}\le {u}_{j}$ is a vector of all zeros except for
± 1$\pm 1$ in position
j$j$, it follows that the working set matrix contains the rows of
$\left(\begin{array}{cc}A& I\end{array}\right)$ and the unit rows associated with the upper and lower bounds in the working set.
The working set matrix
W$W$ can be represented in terms of a certain column partition of the matrix
$\left(\begin{array}{cc}A& I\end{array}\right)$ by (conceptually) partitioning the constraints
Ax − s = 0$Axs=0$ so that
where
B$B$ is a square nonsingular basis and
x_{B}${x}_{B}$,
x_{S}${x}_{S}$ and
x_{N}${x}_{N}$ are the basic, superbasic and nonbasic variables respectively. The nonbasic variables are equal to their upper or lower bounds at
(x,s)$(x,s)$, and the superbasic variables are independent variables that are chosen to improve the value of the current objective function. The number of superbasic variables is
n_{S}${n}_{S}$ (the printed quantity
Ns; see
Section [Description of Monitoring Information]). Given values of
x_{N}${x}_{N}$ and
x_{S}${x}_{S}$, the basic variables
x_{B}${x}_{B}$ are adjusted so that
(x,s)$(x,s)$ satisfies
(5).
If
P$P$ is a permutation matrix such that
P =
$\left(\begin{array}{cc}A& I\end{array}\right)P=\left(\begin{array}{ccc}B& S& N\end{array}\right)$, then
W$W$ satisfies
where
I_{N}${I}_{N}$ is the identity matrix with the same number of columns as
N$N$.
The null space matrix
Z$Z$ is defined from a sparse
LU$LU$ factorization of part of
W$W$. In particular,
Z$Z$ is maintained in ‘reduced gradient’ form, using the LUSOL package (see
Gill et al. (1991)) to maintain sparse
LU$LU$ factors of the basis matrix
B$B$ that alters as the working set
W$W$ changes. Given the permutation
P$P$, the null space basis is given by
This matrix is used only as an operator, i.e., it is never computed explicitly. Products of the form
Zv$Zv$ and
Z^{T}g${Z}^{\mathrm{T}}g$ are obtained by solving with
B$B$ or
B^{T}${B}^{\mathrm{T}}$. This choice of
Z$Z$ implies that
n_{Z}${n}_{Z}$, the number of ‘degrees of freedom’ at
(x,s)$(x,s)$, is the same as
n_{S}${n}_{S}$, the number of superbasic variables.
Let
g_{Z}${g}_{Z}$ and
H_{Z}${H}_{Z}$ denote the
reduced gradient and
reduced Hessian of the objective function:
where
g$g$ is the objective gradient at
(x,s)$(x,s)$. Roughly speaking,
g_{Z}${g}_{Z}$ and
H_{Z}${H}_{Z}$ describe the first and second derivatives of an
n_{S}${n}_{S}$dimensional
unconstrained problem for the calculation of
p_{Z}${p}_{Z}$. (The condition estimator of
H_{Z}${H}_{Z}$ is the quantity
Cond Hz in the monitoring file output; see
Section [Description of Monitoring Information].)
At each iteration, an upper triangular factor R$R$ is available such that H_{Z} = R^{T}R${H}_{Z}={R}^{\mathrm{T}}R$. Normally, R$R$ is computed from R^{T}R = Z^{T}HZ${R}^{\mathrm{T}}R={Z}^{\mathrm{T}}HZ$ at the start of the optimality phase and then updated as the QP working set changes. For efficiency, the dimension of R$R$ should not be excessive (say, n_{S} ≤ 1000${n}_{S}\le 1000$). This is guaranteed if the number of nonlinear variables is ‘moderate’.
If the QP problem contains linear variables,
H$H$ is positive semidefinite and
R$R$ may be singular with at least one zero diagonal element. However, an inertiacontrolling strategy is used to ensure that only the last diagonal element of
R$R$ can be zero. (See
Gill et al. (1991) for a discussion of a similar strategy for indefinite quadratic programming.)
If the initial R$R$ is singular, enough variables are fixed at their current value to give a nonsingular R$R$. This is equivalent to including temporary bound constraints in the working set. Thereafter, R$R$ can become singular only when a constraint is deleted from the working set (in which case no further constraints are deleted until R$R$ becomes nonsingular).
Main Iteration
If the reduced gradient is zero,
(x,s)$(x,s)$ is a constrained stationary point on the working set. During the feasibility phase, the reduced gradient will usually be zero only at a vertex (although it may be zero elsewhere in the presence of constraint dependencies). During the optimality phase, a zero reduced gradient implies that
x$x$ minimizes the quadratic objective function when the constraints in the working set are treated as equalities. At a constrained stationary point, Lagrange multipliers
λ$\lambda $ are defined from the equations
A Lagrange multiplier
λ_{j}${\lambda}_{j}$ corresponding to an inequality constraint in the working set is said to be
optimal if
λ_{j} ≤ σ${\lambda}_{j}\le \sigma $ when the associated constraint is at its
upper bound, or if
λ_{j} ≥ − σ${\lambda}_{j}\ge \sigma $ when the associated constraint is at its
lower bound, where
σ$\sigma $ depends on the value of the optional parameter
Optimality Tolerance. If a multiplier is nonoptimal, the objective function (either the true objective or the sum of infeasibilities) can be reduced by continuing the minimization with the corresponding constraint excluded from the working set. (This step is sometimes referred to as ‘deleting’ a constraint from the working set.) If optimal multipliers occur during the feasibility phase but the sum of infeasibilities is nonzero, there is no feasible point and the function terminates immediately with
ifail = 3${\mathbf{ifail}}={\mathbf{3}}$ (see
Section [Error Indicators and Warnings]).
The special form
(6) of the working set allows the multiplier vector
λ$\lambda $, the solution of
(9), to be written in terms of the vector
where
π$\pi $ satisfies the equations
B^{T}π = g_{B}${B}^{\mathrm{T}}\pi ={g}_{B}$, and
g_{B}${g}_{B}$ denotes the basic elements of
g$g$. The elements of
π$\pi $ are the Lagrange multipliers
λ_{j}${\lambda}_{j}$ associated with the equality constraints
Ax − s = 0$Axs=0$. The vector
d_{N}${d}_{N}$ of nonbasic elements of
d$d$ consists of the Lagrange multipliers
λ_{j}${\lambda}_{j}$ associated with the upper and lower bound constraints in the working set. The vector
d_{S}${d}_{S}$ of superbasic elements of
d$d$ is the reduced gradient
g_{Z}${g}_{Z}$ in
(8). The vector
d_{B}${d}_{B}$ of basic elements of
d$d$ is zero, by construction. (The Euclidean norm of
d_{S}${d}_{S}$ and the final values of
d_{S}${d}_{S}$,
g$g$ and
π$\pi $ are the quantities
Norm rg,
Reduced Gradnt,
Obj Gradient and
Dual Activity in the monitoring file output; see
Section [Description of Monitoring Information].)
If the reduced gradient is not zero, Lagrange multipliers need not be computed and the search direction is given by
p = Zp_{Z}$p=Z{p}_{Z}$ (see
(7) and
(11)). The step length is chosen to maintain feasibility with respect to the satisfied constraints.
There are two possible choices for
p_{Z}${p}_{Z}$, depending on whether or not
H_{Z}${H}_{Z}$ is singular. If
H_{Z}${H}_{Z}$ is nonsingular,
R$R$ is nonsingular and
p_{Z}${p}_{Z}$ in
(4) is computed from the equations
where
g_{Z}${g}_{Z}$ is the reduced gradient at
x$x$. In this case,
(x,s) + p$(x,s)+p$ is the minimizer of the objective function subject to the working set constraints being treated as equalities. If
(x,s) + p$(x,s)+p$ is feasible,
α$\alpha $ is defined to be unity. In this case, the reduced gradient at
(x,s)$(\stackrel{}{x},\stackrel{}{s})$ will be zero, and Lagrange multipliers are computed at the next iteration. Otherwise,
α$\alpha $ is set to
α_{m}${\alpha}_{{\mathbf{m}}}$, the step to the ‘nearest’ constraint along
p$p$. This constraint is then added to the working set at the next iteration.
If
H_{Z}${H}_{Z}$ is singular, then
R$R$ must also be singular, and an inertiacontrolling strategy is used to ensure that only the last diagonal element of
R$R$ is zero. (See
Gill et al. (1991) for a discussion of a similar strategy for indefinite quadratic programming.) In this case,
p_{Z}${p}_{Z}$ satisfies
which allows the objective function to be reduced by any step of the form
(x,s) + αp$(x,s)+\alpha p$,
where
α > 0$\alpha >0$. The vector
p = Zp_{Z}$p=Z{p}_{Z}$ is a direction of unbounded descent for the QP problem in the sense that the QP
objective is linear and decreases without bound along
p$p$. If no finite step of the form
(x,s) + αp$(x,s)+\alpha p$ (where
α > 0$\alpha >0$) reaches a constraint not in the working set, the QP problem is unbounded and the function terminates immediately with
ifail = 2${\mathbf{ifail}}={\mathbf{2}}$ (see
Section [Error Indicators and Warnings]). Otherwise,
α$\alpha $ is defined as the maximum feasible step along
p$p$ and a constraint active at
(x,s) + αp$(x,s)+\alpha p$ is added to the working set for the next iteration.
Miscellaneous
If the basis matrix is not chosen carefully, the condition of the null space matrix
Z$Z$ in
(7) could be arbitrarily high. To guard against this, the function implements a ‘basis repair’ feature in which the LUSOL package (see
Gill et al. (1991)) is used to compute the rectangular factorization
returning just the permutation
P$P$ that makes
PLP^{T}$PL{P}^{\mathrm{T}}$ unit lower triangular. The pivot tolerance is set to require
PLP^{T}_{ij} ≤ 2${\leftPL{P}^{\mathrm{T}}\right}_{ij}\le 2$, and the permutation is used to define
P$P$ in
(6). It can be shown that
‖Z‖$\Vert Z\Vert $ is likely to be little more than unity. Hence,
Z$Z$ should be wellconditioned
regardless of the condition of
W$W$. This feature is applied at the beginning of the optimality phase if a potential
B − S$BS$ ordering is known.
The EXPAND procedure (see
Gill et al. (1989)) is used to reduce the possibility of cycling at a point where the active constraints are nearly linearly dependent. Although there is no absolute guarantee that cycling will not occur, the probability of cycling is extremely small (see
Hall and McKinnon (1996)). The main feature of EXPAND is that the feasibility tolerance is increased at the start of every iteration. This allows a positive step to be taken at every iteration, perhaps at the expense of violating the bounds on
(x,s)$(x,s)$ by a small amount.
Suppose that the value of the optional parameter
Feasibility Tolerance is
δ$\delta $. Over a period of
K$K$ iterations (where
K$K$ is the value of the optional parameter
Expand Frequency), the feasibility tolerance actually used by the function (i.e., the
working feasibility tolerance) increases from
0.5δ$0.5\delta $ to
δ$\delta $ (in steps of
0.5δ / K$0.5\delta /K$).
At certain stages the following ‘resetting procedure’ is used to remove small constraint infeasibilities. First, all nonbasic variables are moved exactly onto their bounds. A count is kept of the number of nontrivial adjustments made. If the count is nonzero, the basic variables are recomputed. Finally, the working feasibility tolerance is reinitialized to 0.5δ$0.5\delta $.
If a problem requires more than K$K$ iterations, the resetting procedure is invoked and a new cycle of iterations is started. (The decision to resume the feasibility phase or optimality phase is based on comparing any constraint infeasibilities with δ$\delta $.)
The resetting procedure is also invoked when the function reaches an apparently optimal, infeasible or unbounded solution, unless this situation has already occurred twice. If any nontrivial adjustments are made, iterations are continued.
The EXPAND procedure not only allows a positive step to be taken at every iteration, but also provides a potential choice of constraints to be added to the working set. All constraints at a distance α$\alpha $ (where α ≤ α_{m}$\alpha \le {\alpha}_{{\mathbf{m}}}$) along p$p$ from the current point are then viewed as acceptable candidates for inclusion in the working set. The constraint whose normal makes the largest angle with the search direction is added to the working set. This strategy helps keep the basis matrix B$B$ wellconditioned.
Optional Parameters
Several optional parameters in nag_opt_qpconvex1_sparse_solve (e04nk) define choices in the problem specification or the algorithm logic. In order to reduce the number of formal parameters of nag_opt_qpconvex1_sparse_solve (e04nk) these optional parameters have associated default values that are appropriate for most problems. Therefore, you need only specify those optional parameters whose values are to be different from their default values.
The remainder of this section can be skipped if you wish to use the default values for all optional parameters.
The following is a list of the optional parameters available. A full description of each optional parameter is provided in
Section [Description of the Optional s].
Optional parameters may be specified by calling
nag_opt_qpconvex1_sparse_option_string (e04nm) before a call to
nag_opt_qpconvex1_sparse_solve (e04nk).
nag_opt_qpconvex1_sparse_option_string (e04nm) can be called to supply options directly, one call being necessary for each optional parameter. For example,
[lwsav, iwsav, rwsav, inform] = e04nm('Print Level = 5', lwsav, iwsav, rwsav);
nag_opt_qpconvex1_sparse_option_string (e04nm) should be consulted for a full description of this method of supplying optional parameters.
All optional parameters not specified by you are set to their default values. Optional parameters specified by you are unaltered by nag_opt_qpconvex1_sparse_solve (e04nk) (unless they define invalid values) and so remain in effect for subsequent calls unless altered by you.
Description of the Optional Parameters
For each option, we give a summary line, a description of the optional parameter and details of constraints.
The summary line contains:
 the keywords, where the minimum abbreviation of each keyword is underlined (if no characters of an optional qualifier are underlined, the qualifier may be omitted);
 a parameter value,
where the letters a$a$, i and r$i\text{ and}r$ denote options that take character, integer and real values respectively;
 the default value is used whenever the condition i ≥ 100000000$\lefti\right\ge 100000000$ is satisfied and where the symbol ε$\epsilon $ is a generic notation for machine precision (see nag_machine_precision (x02aj));
Keywords and character values are case and white space insensitive.
Check Frequency i$i$Default = 60$\text{}=60$Every i$i$th iteration after the most recent basis factorization, a numerical test is made to see if the current solution (x,s)$(x,s)$ satisfies the linear constraints Ax − s = 0$Axs=0$. If the largest element of the residual vector r = Ax − s$r=Axs$ is judged to be too large, the current basis is refactorized and the basic variables recomputed to satisfy the constraints more accurately. If i < 0$i<0$, the default value is used. If i = 0$i=0$, the value i = 99999999$i=99999999$ is used and effectively no checks are made.
Crash Option i$i$Default = 2$\text{}=2$Note that this option does not apply when
start = 'W'${\mathbf{start}}=\text{'W'}$ (see
Section [Parameters]).
If
start = 'C'${\mathbf{start}}=\text{'C'}$, an internal Crash procedure is used to select an initial basis from various rows and columns of the constraint matrix
$\left(\begin{array}{cc}A& I\end{array}\right)$. The value of
i$i$ determines which rows and columns are initially eligible for the basis, and how many times the Crash procedure is called. If
i = 0$i=0$, the allslack basis
B = − I$B=I$ is chosen. If
i = 1$i=1$, the Crash procedure is called once (looking for a triangular basis in all rows and columns of the linear constraint matrix
A$A$). If
i = 2$i=2$, the Crash procedure is called twice (looking at any
equality constraints first followed by any
inequality constraints). If
i < 0$i<0$ or
i > 2$i>2$, the default value is used.
If i = 1 or 2$i=1\text{ or}2$, certain slacks on inequality rows are selected for the basis first. (If i = 2$i=2$, numerical values are used to exclude slacks that are close to a bound.) The Crash procedure then makes several passes through the columns of A$A$, searching for a basis matrix that is essentially triangular. A column is assigned to ‘pivot’ on a particular row if the column contains a suitably large element in a row that has not yet been assigned. (The pivot elements ultimately form the diagonals of the triangular basis.) For remaining unassigned rows, slack variables are inserted to complete the basis.
Crash Tolerance r$r$Default = 0.1$\text{}=0.1$This value allows the Crash procedure to ignore certain ‘small’ nonzero elements in the constraint matrix A$A$ while searching for a triangular basis. For each column of A$A$, if a_{max}${a}_{\mathrm{max}}$ is the largest element in the column, other nonzeros in that column are ignored if they are less than (or equal to) a_{max} × r${a}_{\mathrm{max}}\times r$.
When r > 0$r>0$, the basis obtained by the Crash procedure may not be strictly triangular, but it is likely to be nonsingular and almost triangular. The intention is to obtain a starting basis with more column variables and fewer (arbitrary) slacks. A feasible solution may be reached earlier for some problems. If r < 0$r<0$ or r ≥ 1$r\ge 1$, the default value is used.
Defaults This special keyword may be used to reset all optional parameters to their default values.
Expand Frequency i$i$Default = 10000$\text{}=10000$This option is part of an anticycling procedure (see
Section [Miscellaneous]) designed to allow progress even on highly degenerate problems.
For LP problems, the strategy is to force a positive step at every iteration, at the expense of violating the constraints by a small amount. Suppose that the value of the optional parameter
Feasibility Tolerance is
δ$\delta $. Over a period of
i$i$ iterations, the feasibility tolerance actually used by
nag_opt_qpconvex1_sparse_solve (e04nk) (i.e., the
working feasibility tolerance) increases from
0.5δ$0.5\delta $ to
δ$\delta $ (in steps of
0.5δ / i$0.5\delta /i$).
For QP problems, the same procedure is used for iterations in which there is only one superbasic variable. (Cycling can only occur when the current solution is at a vertex of the feasible region.) Thus, zero steps are allowed if there is more than one superbasic variable, but otherwise positive steps are enforced.
Increasing the value of
i$i$ helps reduce the number of slightly infeasible nonbasic basic variables (most of which are eliminated during the resetting procedure). However, it also diminishes the freedom to choose a large pivot element (see optional parameter
Pivot Tolerance).
If i < 0$i<0$, the default value is used. If i = 0$i=0$, the value i = 99999999$i=99999999$ is used and effectively no anticycling procedure is invoked.
Factorization Frequency i$i$Default = 100$\text{}=100$If
i > 0$i>0$, at most
i$i$ basis changes will occur between factorizations of the basis matrix. For LP problems, the basis factors are usually updated at every iteration. For QP problems, fewer basis updates will occur as the solution is approached. The number of iterations between basis factorizations will therefore increase. During these iterations a test is made regularly according to the value of optional parameter
Check Frequency to ensure that the linear constraints
Ax − s = 0$Axs=0$ are satisfied. If necessary, the basis will be refactorized before the limit of
i$i$ updates is reached. If
i ≤ 0$i\le 0$, the default value is used.
Feasibility Tolerance r$r$Default = max (10^{ − 6},sqrt(ε))$\text{}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}({10}^{6},\sqrt{\epsilon})$If r ≥ ε$r\ge \epsilon $, r$r$ defines the maximum acceptable absolute violation in each constraint at a ‘feasible’ point (including slack variables). For example, if the variables and the coefficients in the linear constraints are of order unity, and the latter are correct to about five decimal digits, it would be appropriate to specify r$r$ as 10^{ − 5}${10}^{5}$. If r < ε$r<\epsilon $, the default value is used.
nag_opt_qpconvex1_sparse_solve (e04nk) attempts to find a feasible solution before optimizing the objective function. If the sum of infeasibilities cannot be reduced to zero, the problem is assumed to be infeasible. Let Sinf be the corresponding sum of infeasibilities. If Sinf is quite small, it may be appropriate to raise r$r$ by a factor of 10$10$ or 100$100$. Otherwise, some error in the data should be suspected. Note that the function does not attempt to find the minimum value of Sinf.
If the constraints and variables have been scaled (see
Scale Option), then feasibility is defined in terms of the scaled problem (since it is more likely to be meaningful).
Infinite Bound Size r$r$Default = 10^{20}$\text{}={10}^{20}$If r > 0$r>0$, r$r$ defines the ‘infinite’ bound bigbnd$\mathit{bigbnd}$ in the definition of the problem constraints. Any upper bound greater than or equal to bigbnd$\mathit{bigbnd}$ will be regarded as + ∞$+\infty $ (and similarly any lower bound less than or equal to − bigbnd$\mathit{bigbnd}$ will be regarded as − ∞$\infty $). If r ≤ 0$r\le 0$, the default value is used.
Infinite Step Size r$r$Default = max (bigbnd,10^{20})$\text{}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}(\mathit{bigbnd},{10}^{20})$If r > 0$r>0$, r$r$ specifies the magnitude of the change in variables that will be considered a step to an unbounded solution. (Note that an unbounded solution can occur only when the Hessian is not positive definite.) If the change in x$x$ during an iteration would exceed the value of r$r$, the objective function is considered to be unbounded below in the feasible region. If r ≤ 0$r\le 0$, the default value is used.
Iteration Limit i$i$Default = max (50,5(n + m))$\text{}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}(50,5(n+m))$Iters Itns The value of
i$i$ specifies the maximum number of iterations allowed before termination. Setting
i = 0$i=0$ and
Print Level > 0${\mathbf{Print\; Level}}>0$ means that the workspace needed to start solving the problem will be computed and printed, but no iterations will be performed. If
i < 0$i<0$, the default value is used.
List Default for e04nk = List$\text{e04nk}={\mathbf{List}}$Nolist Default for e04nk = Nolist$\text{e04nk}={\mathbf{Nolist}}$Normally each optional parameter specification is printed as it is supplied. Optional parameter
Nolist may be used to suppress the printing and optional parameter
List may be used to restore printing.
LU Factor Tolerance r_{1}${r}_{1}$Default = 100.0$\text{}=100.0$LU Update Tolerance r_{2}${r}_{2}$Default = 10.0$\text{}=10.0$The values of
r_{1}${r}_{1}$ and
r_{2}${r}_{2}$ affect the stability and sparsity of the basis factorization
B = LU$B=LU$, during refactorization and updates respectively. The lower triangular matrix
L$L$ is a product of matrices of the form
where the multipliers
μ$\mu $ will satisfy
μ ≤ r_{i}$\left\mu \right\le {r}_{i}$. The default values of
r_{1}${r}_{1}$ and
r_{2}${r}_{2}$ usually strike a good compromise between stability and sparsity. For large and relatively dense problems, setting
r_{1}${r}_{1}$ and
r_{2}${r}_{2}$ to
25$25$ (say) may give a marked improvement in sparsity without impairing stability to a serious degree.
Note that for band matrices it may be necessary to set r_{1}${r}_{1}$ in the range 1 ≤ r_{1} < 2$1\le {r}_{1}<2$ in order to achieve stability. If r_{1} < 1${r}_{1}<1$ or r_{2} < 1${r}_{2}<1$, the default value is used.
LU Singularity Tolerance r$r$Default = ε^{0.67}$\text{}={\epsilon}^{0.67}$If r > 0$r>0$, r$r$ defines the singularity tolerance used to guard against illconditioned basis matrices. Whenever the basis is refactorized, the diagonal elements of U$U$ are tested as follows. If u_{jj} ≤ r$\left{u}_{jj}\right\le r$ or u_{jj} < r × max_{i} u_{ij}$\left{u}_{jj}\right<r\times {\displaystyle \underset{i}{\mathrm{max}}}\phantom{\rule{0.25em}{0ex}}\left{u}_{ij}\right$, the j$j$th column of the basis is replaced by the corresponding slack variable. If r ≤ 0$r\le 0$, the default value is used.
Minimize DefaultMaximize This option specifies the required direction of the optimization. It applies to both linear and nonlinear terms (if any) in the objective function. Note that if two problems are the same except that one minimizes
f(x)$f\left(x\right)$ and the other maximizes
− f(x)$f\left(x\right)$, their solutions will be the same but the signs of the dual variables
π_{i}${\pi}_{i}$ and the reduced gradients
d_{j}${d}_{j}$ (see
Section [Main Iteration]) will be reversed.
Monitoring File i$i$Default = − 1$\text{}=1$If
i ≥ 0$i\ge 0$ and
Print Level > 0${\mathbf{Print\; Level}}>0$ (see
Print Level), monitoring information produced by
nag_opt_qpconvex1_sparse_solve (e04nk) is sent to a file with logical unit number
i$i$. If
i < 0$i<0$ and/or
Print Level = 0${\mathbf{Print\; Level}}=0$, the default value is used and hence no monitoring information is produced.
Optimality Tolerance r$r$Default = max (10^{ − 6},sqrt(ε))$\text{}=\mathrm{max}\phantom{\rule{0.125em}{0ex}}({10}^{6},\sqrt{\epsilon})$If r ≥ ε$r\ge \epsilon $, r$r$ is used to judge the size of the reduced gradients d_{j} = g_{j} − π^{T}a_{j}${d}_{j}={g}_{j}{\pi}^{\mathrm{T}}{a}_{j}$. By definition, the reduced gradients for basic variables are always zero. Optimality is declared if the reduced gradients for any nonbasic variables at their lower or upper bounds satisfy − r × max (1,‖π‖) ≤ d_{j} ≤ r × max (1,‖π‖)$r\times \mathrm{max}\phantom{\rule{0.125em}{0ex}}(1,\Vert \pi \Vert )\le {d}_{j}\le r\times \mathrm{max}\phantom{\rule{0.125em}{0ex}}(1,\Vert \pi \Vert )$, and if d_{j} ≤ r × max (1,‖π‖)$\left{d}_{j}\right\le r\times \mathrm{max}\phantom{\rule{0.125em}{0ex}}(1,\Vert \pi \Vert )$ for any superbasic variables. If r < ε$r<\epsilon $, the default value is used.
Partial Price i$i$Default = 10$\text{}=10$Note that this option does not apply to QP problems.
This option is recommended for large FP or LP problems that have significantly more variables than constraints (i.e.,
n ≫ m$n\gg m$). It reduces the work required for each pricing operation (i.e., when a nonbasic variable is selected to enter the basis). If
i = 1$i=1$, all columns of the constraint matrix
$\left(\begin{array}{cc}A& I\end{array}\right)$ are searched. If
i > 1$i>1$,
A$A$ and
− I$I$ are partitioned to give
i$i$ roughly equal segments
A_{j},K_{j}${A}_{\mathit{j}},{K}_{\mathit{j}}$, for
j = 1,2, … ,p$\mathit{j}=1,2,\dots ,p$ (modulo
p$p$). If the previous pricing search was successful on
A_{j − 1},K_{j − 1}${A}_{j1},{K}_{j1}$, the next search begins on the segments
A_{j},K_{j}${A}_{j},{K}_{j}$. If a reduced gradient is found that is larger than some dynamic tolerance, the variable with the largest such reduced gradient (of appropriate sign) is selected to enter the basis. If nothing is found, the search continues on the next segments
A_{j + 1},K_{j + 1}${A}_{j+1},{K}_{j+1}$, and so on. If
i ≤ 0$i\le 0$, the default value is used.
Pivot Tolerance r$r$Default = ε^{0.67}$\text{}={\epsilon}^{0.67}$If r > 0$r>0$, r$r$ is used to prevent columns entering the basis if they would cause the basis to become almost singular. If r ≤ 0$r\le 0$, the default value is used.
Print Level i$i$The value of
i$i$ controls the amount of printout produced by
nag_opt_qpconvex1_sparse_solve (e04nk), as indicated below. A detailed description of the printed output is given in
Section [Printed output] (summary output at each iteration and the final solution) and
Section [Description of Monitoring Information] (monitoring information at each iteration). Note that the summary output will not exceed
80$80$ characters per line and that the monitoring information will not exceed
120$120$ characters per line. If
i < 0$i<0$, the default value is used.
The following printout is sent to the current advisory message unit (as defined by
nag_file_set_unit_advisory (x04ab)):
i$i$ 
Output 
≥ 00$\phantom{\ge 0}0$ 
No output. 
≥ 01$\phantom{\ge 0}1$ 
The final solution only. 
≥ 05$\phantom{\ge 0}5$ 
One line of summary output for each iteration (no printout of the final solution). 
≥ 10$\text{}\ge 10$ 
The final solution and one line of summary output for each iteration. 
The following printout is sent to the logical unit number defined by the optional parameter
Monitoring File:
i$i$ 
Output 
≥ 00$\phantom{\ge 0}0$ 
No output. 
≥ 01$\phantom{\ge 0}1$ 
The final solution only. 
≥ 05$\phantom{\ge 0}5$ 
One long line of output for each iteration (no printout of the final solution). 
≥ 10$\text{}\ge 10$ 
The final solution and one long line of output for each iteration. 
≥ 20$\text{}\ge 20$ 
The final solution, one long line of output for each iteration, matrix statistics (initial status of rows and columns, number of elements, density, biggest and smallest elements, etc.), details of the scale factors resulting from the scaling procedure (if Scale Option = 1${\mathbf{Scale\; Option}}=1$ or 2$2$ (see the description of the optional parameter Scale Option), basis factorization statistics and details of the initial basis resulting from the Crash procedure (if start = 'C'${\mathbf{start}}=\text{'C'}$; see Section [Parameters]). 
If
Print Level > 0${\mathbf{Print\; Level}}>0$ and the unit number defined by optional parameter
Monitoring File is the same as that defined by
nag_file_set_unit_advisory (x04ab), then the summary output is suppressed.
Rank Tolerance r$r$Default = 100ε$\text{}=100\epsilon $Scale Option i$i$Default = 2$\text{}=2$This option enables you to scale the variables and constraints using an iterative procedure due to
Fourer (1982), which attempts to compute row scales
r_{i}${r}_{i}$ and column scales
c_{j}${c}_{j}$ such that the scaled matrix coefficients
a_{ij} = a_{ij} × (c_{j} / r_{i})${\stackrel{}{a}}_{ij}={a}_{ij}\times ({c}_{j}/{r}_{i})$ are as close as possible to unity. This may improve the overall efficiency on some problems. (The lower and upper bounds on the variables and slacks for the scaled problem are redefined as
l_{j} = l_{j} / c_{j}${\stackrel{}{l}}_{j}={l}_{j}/{c}_{j}$ and
u_{j} = u_{j} / c_{j}${\stackrel{}{u}}_{j}={u}_{j}/{c}_{j}$ respectively, where
c_{j} ≡ r_{j − n}${c}_{j}\equiv {r}_{jn}$ if
j > n$j>n$.)
If
i = 0$i=0$, no scaling is performed. If
i = 1$i=1$, all rows and columns of the constraint matrix
A$A$ are scaled. If
i = 2$i=2$, an additional scaling is performed that may be helpful when the solution
x$x$ is large; it takes into account columns of
$\left(\begin{array}{cc}A& I\end{array}\right)$ that are fixed or have positive lower bounds or negative upper bounds. If
i < 0$i<0$ or
i > 2$i>2$, the default value is used.
Scale Tolerance r$r$Default = 0.9$\text{}=0.9$Note that this option does not apply when
Scale Option = 0${\mathbf{Scale\; Option}}=0$.
If 0 < r < 1$0<r<1$, r$r$ is used to control the number of scaling passes to be made through the constraint matrix A$A$. At least 3$3$ (and at most 10$10$) passes will be made. More precisely, let a_{p}${a}_{p}$ denote the largest column ratio (i.e., ('biggest' element)/('smallest' element)
$\frac{\text{'biggest'}\text{ element}}{\text{'smallest'}\text{ element}}$ in some sense) after the p$p$th scaling pass through A$A$. The scaling procedure is terminated if a_{p} ≥ a_{p − 1} × r${a}_{p}\ge {a}_{p1}\times r$ for some p ≥ 3$p\ge 3$. Thus, increasing the value of r$r$ from 0.9$0.9$ to 0.99$0.99$ (say) will probably increase the number of passes through A$A$.
If r ≤ 0$r\le 0$ or r ≥ 1$r\ge 1$, the default value is used.
Superbasics Limit i$i$Default = min (n_{H} + 1,n)$\text{}=\mathrm{min}\phantom{\rule{0.125em}{0ex}}({n}_{H}+1,n)$Note that this option does not apply to FP or LP problems.
The value of i$i$ specifies ‘how nonlinear’ you expect the QP problem to be. If i ≤ 0$i\le 0$, the default value is used.
Description of Monitoring Information
This section describes the intermediate printout and final printout which constitutes the monitoring information produced by
nag_opt_qpconvex1_sparse_solve (e04nk). (See also the description of the optional parameters
Monitoring File and
Print Level.) You can control the level of printed output.
When
Print Level = 5${\mathbf{Print\; Level}}=5$ or
≥ 10$\text{}\ge 10$ and
Monitoring File ≥ 0${\mathbf{Monitoring\; File}}\ge 0$, the following line of intermediate printout (
< 120$\text{}<120$ characters) is produced at every iteration on the unit number specified by optional parameter
Monitoring File. Unless stated otherwise, the values of the quantities printed are those in effect
on
completion of the given iteration.
Itn 
is the iteration count.

pp 
is the partial price indicator. The variable selected by the last pricing operation came from the ppth partition of A$A$ and − I$I$. Note that pp is reset to zero whenever the basis is refactorized.

dj 
is the value of the reduced gradient (or reduced cost) for the variable selected by the pricing operation at the start of the current iteration.

+S 
is the variable selected by the pricing operation to be added to the superbasic set.

S 
is the variable chosen to leave the superbasic set.

BS 
is the variable removed from the basis (if any) to become nonbasic.

Step 
is the value of the step length α$\alpha $ taken along the current search direction p$p$. The variables x$x$ have just been changed to x + αp$x+\alpha p$. If a variable is made superbasic during the current iteration (i.e., +S is positive), Step will be the step to the nearest bound. During the optimality phase, the step can be greater than unity only if the reduced Hessian is not positive definite.

Pivot 
is the r$r$th element of a vector y$y$ satisfying By = a_{q}$By={a}_{q}$ whenever a_{q}${a}_{q}$ (the q$q$th column of the constraint matrix
$\left(\begin{array}{cc}A& I\end{array}\right)$) replaces the r$r$th column of the basis matrix B$B$. Wherever possible, Step is chosen so as to avoid extremely small values of Pivot (since they may cause the basis to be nearly singular). In extreme cases, it may be necessary to increase the value of the optional parameter Pivot Tolerance to exclude very small elements of y$y$ from consideration during the computation of Step.

Ninf 
is the number of violated constraints (infeasibilities). This will be zero during the optimality phase.

Sinf/Objective 
is the value of the current objective function. If x$x$ is not feasible, Sinf gives the sum of the magnitudes of constraint violations. If x$x$ is feasible, Objective is the value of the objective function. The output line for the final iteration of the feasibility phase (i.e., the first iteration for which Ninf is zero) will give the value of the true objective at the first feasible point. During the optimality phase, the value of the objective function will be nonincreasing. During the feasibility phase, the number of constraint infeasibilities will not increase until either a feasible point is found, or the optimality of the multipliers implies that no feasible point exists.

L 
is the number of nonzeros in the basis factor L$L$. Immediately after a basis factorization B = LU$B=LU$, this entry contains lenL. Further nonzeros are added to L when various columns of B$B$ are later replaced. (Thus, L increases monotonically.)

U 
is the number of nonzeros in the basis factor U$U$. Immediately after a basis factorization B = LU$B=LU$, this entry contains lenU. As columns of B$B$ are replaced, the matrix U$U$ is maintained explicitly (in sparse form). The value of U may fluctuate up or down; in general, it will tend to increase.

Ncp 
is the number of compressions required to recover workspace in the data structure for U$U$. This includes the number of compressions needed during the previous basis factorization. Normally, Ncp should increase very slowly. If it does not, increase leniz and lenz by at least L + U$\mathtt{L}+\mathtt{U}$ and rerun nag_opt_qpconvex1_sparse_solve (e04nk) (possibly using start = 'W'${\mathbf{start}}=\text{'W'}$; see Section [Parameters]).

Norm rg 
is ‖d_{S}‖$\Vert {d}_{S}\Vert $, the Euclidean norm of the reduced gradient (see Section [Main Iteration]). During the optimality phase, this norm will be approximately zero after a unit step. For FP and LP problems, Norm rg is not printed.

Ns 
is the current number of superbasic variables. For FP and LP problems, Ns is not printed.

Cond Hz 
is a lower bound on the condition number of the reduced Hessian (see Section [Definition of the Working Set and Search Direction]). The larger this number, the more difficult the problem. For FP and LP problems, Cond Hz is not printed.

When
Print Level ≥ 20${\mathbf{Print\; Level}}\ge 20$ and
Monitoring File ≥ 0${\mathbf{Monitoring\; File}}\ge 0$, the following lines of intermediate printout (
< 120$\text{}<120$ characters) are produced on the unit number specified by optional parameter
Monitoring File whenever the matrix
B$B$ or
BS = (BS)${B}_{S}={\left(\begin{array}{cc}B& S\end{array}\right)}^{\mathrm{T}}$ is factorized. Gaussian elimination is used to compute an
LU$LU$ factorization of
B$B$ or
B_{S}${B}_{S}$, where
PLP^{T}$PL{P}^{\mathrm{T}}$ is a lower triangular matrix and
PUQ$PUQ$ is an upper triangular matrix for some permutation matrices
P$P$ and
Q$Q$. The factorization is stabilized in the manner described under the optional parameter
LU Factor Tolerance (
default value = 100.0$\text{default value}=100.0$).
Factorize 
is the factorization count.

Demand 
is a code giving the reason for the present factorization as follows:
Code 
Meaning 
010$\phantom{01}0$ 
First LU$LU$ factorization. 
011$\phantom{01}1$ 
The number of updates reached the value of the optional parameter Factorization Frequency. 
012$\phantom{01}2$ 
The number of nonzeros in the updated factors has increased significantly. 
017$\phantom{01}7$ 
Not enough storage to update factors. 
010$\phantom{0}10$ 
Row residuals too large (see the description for the optional parameter Check Frequency). 
011$\phantom{0}11$ 
Illconditioning has caused inconsistent results. 

Iteration 
is the iteration count.

Nonlinear 
is the number of nonlinear variables in the current basis B$B$ (not printed if B_{S}${B}_{S}$ is factorized).

Linear 
is the number of linear variables in B$B$ (not printed if B_{S}${B}_{S}$ is factorized).

Slacks 
is the number of slack variables in B$B$ (not printed if B_{S}${B}_{S}$ is factorized).

Elems 
is the number of nonzeros in B$B$ (not printed if B_{S}${B}_{S}$ is factorized).

Density 
is the percentage nonzero density of B$B$ (not printed if B_{S}${B}_{S}$ is factorized). More precisely, Density = 100 × Elems / (Nonlinear + Linear + Slacks)^{2}$\mathtt{Density}=100\times \mathtt{Elems}/{(\mathtt{Nonlinear}+\mathtt{Linear}+\mathtt{Slacks})}^{2}$.

Compressns 
is the number of times the data structure holding the partially factorized matrix needed to be compressed, in order to recover unused workspace. Ideally, it should be zero. If it is more than 3$3$ or 4$4$, increase leniz and lenz and rerun nag_opt_qpconvex1_sparse_solve (e04nk) (possibly using start = 'W'${\mathbf{start}}=\text{'W'}$; see Section [Parameters]).

Merit 
is the average Markowitz merit count for the elements chosen to be the diagonals of PUQ$PUQ$. Each merit count is defined to be (c − 1)(r − 1)$(c1)(r1)$, where c$c$ and r$r$ are the number of nonzeros in the column and row containing the element at the time it is selected to be the next diagonal. Merit is the average of m such quantities. It gives an indication of how much work was required to preserve sparsity during the factorization.

lenL 
is the number of nonzeros in L$L$.

lenU 
is the number of nonzeros in U$U$.

Increase 
is the percentage increase in the number of nonzeros in L$L$ and U$U$ relative to the number of nonzeros in B$B$. More precisely, Increase = 100 × (lenL + lenU − Elems) / Elems$\mathtt{Increase}=100\times (\mathtt{lenL}+\mathtt{lenU}\mathtt{Elems})/\mathtt{Elems}$.

m 
is the number of rows in the problem. Note that m = Ut + Lt + bp$\mathtt{m}=\mathtt{Ut}+\mathtt{Lt}+\mathtt{bp}$.

Ut 
is the number of triangular rows of B$B$ at the top of U$U$.

d1 
is the number of columns remaining when the density of the basis matrix being factorized reached 0.3$0.3$.

Lmax 
is the maximum subdiagonal element in the columns of L$L$. This will not exceed the value of the optional parameter LU Factor Tolerance.

Bmax 
is the maximum nonzero element in B$B$ (not printed if B_{S}${B}_{S}$ is factorized).

BSmax 
is the maximum nonzero element in B_{S}${B}_{S}$ (not printed if B$B$ is factorized).

Umax 
is the maximum nonzero element in U$U$, excluding elements of B$B$ that remain in U$U$ unchanged. (For example, if a slack variable is in the basis, the corresponding row of B$B$ will become a row of U$U$ without modification. Elements in such rows will not contribute to Umax. If the basis is strictly triangular then none of the elements of B$B$ will contribute and Umax will be zero.)Ideally, Umax should not be significantly larger than Bmax. If it is several orders of magnitude larger, it may be advisable to reset the optional parameter LU Factor Tolerance to some value nearer unity. Umax is not printed if B_{S}${B}_{S}$ is factorized.

Umin 
is the magnitude of the smallest diagonal element of PUQ$PUQ$ (not printed if B_{S}${B}_{S}$ is factorized).

Growth 
is the value of the ratio Umax/Bmax, which should not be too large. Providing Lmax is not large (say, < 10.0$\text{}<10.0$), the ratio max (Bmax,Umax) / Umin$\mathrm{max}\phantom{\rule{0.125em}{0ex}}(\mathtt{Bmax},\mathtt{Umax})/\mathtt{Umin}$ is an estimate of the condition number of B$B$. If this number is extremely large, the basis is nearly singular and some numerical difficulties might occur. (However, an effort is made to avoid nearsingularity by using slacks to replace columns of B$B$ that would have made Umin extremely small and the modified basis is refactorized.) Growth is not printed if B_{S}${B}_{S}$ is factorized. 
Lt 
is the number of triangular columns of B$B$ at the left of L$L$.

bp 
is the size of the ‘bump’ or block to be factorized nontrivially after the triangular rows and columns of B$B$ have been removed.

d2 
is the number of columns remaining when the density of the basis matrix being factorized has reached 0.6$0.6$.

When
Print Level ≥ 20${\mathbf{Print\; Level}}\ge 20$ and
Monitoring File ≥ 0${\mathbf{Monitoring\; File}}\ge 0$, the following lines of intermediate printout (
< 80$\text{}<80$ characters) are produced on the unit number specified by optional parameter
Monitoring File whenever
start = 'C'${\mathbf{start}}=\text{'C'}$ (see
Section [Parameters]). They refer to the number of columns selected by the Crash procedure during each of several passes through
A$A$, whilst searching for a triangular basis matrix.
Slacks 
is the number of slacks selected initially.

Free cols 
is the number of free columns in the basis, including those whose bounds are rather far apart.

Preferred 
is the number of ‘preferred’ columns in the basis (i.e., istate(j) = 3${\mathbf{istate}}\left(j\right)=3$ for some j ≤ n$j\le n$). It will be a subset of the columns for which istate(j) = 3${\mathbf{istate}}\left(j\right)=3$ was specified.

Unit 
is the number of unit columns in the basis.

Double 
is the number of double columns in the basis.

Triangle 
is the number of triangular columns in the basis.

Pad 
is the number of slacks used to pad the basis (to make it a nonsingular triangle).

When
Print Level ≥ 20${\mathbf{Print\; Level}}\ge 20$ and
Monitoring File ≥ 0${\mathbf{Monitoring\; File}}\ge 0$, the following lines of intermediate printout (
< 80$\text{}<80$ characters) are produced on the unit number specified by optional parameter
Monitoring File. They refer to the elements of the
names array (see
Section [Parameters]).
Name 
gives the name for the problem (blank if problem unnamed).

Status 
gives the exit status for the problem (i.e., Optimal soln, Weak soln, Unbounded, Infeasible, Excess itns, Error condn or Feasble soln) followed by details of the direction of the optimization (i.e., (Min) or (Max)).

Objective 
gives the name of the free row for the problem (blank if objective unnamed).

RHS 
gives the name of the constraint righthand side for the problem (blank if objective unnamed).

Ranges 
gives the name of the ranges for the problem (blank if objective unnamed).

Bounds 
gives the name of the bounds for the problem (blank if objective unnamed).

When
Print Level = 1${\mathbf{Print\; Level}}=1$ or
≥ 10$\text{}\ge 10$ and
Monitoring File ≥ 0${\mathbf{Monitoring\; File}}\ge 0$, the following lines of final printout (
< 120$\text{}<120$ characters) are produced on the unit number specified by optional parameter
Monitoring File.
Let
a_{j}${a}_{\mathit{j}}$ denote the
j$\mathit{j}$th column of
A$A$, for
j = 1,2, … ,n$\mathit{j}=1,2,\dots ,n$. The following describes the printout for each column (or variable). A full stop (.) is printed for any numerical value that is zero.
Number 
is the column number j$j$. (This is used internally to refer to x_{j}${x}_{j}$ in the intermediate output.)

Column 
gives the name of x_{j}${x}_{j}$.

State 
gives the state of the variable (LL if nonbasic on its lower bound, UL if nonbasic on its upper bound, EQ if nonbasic and fixed, FR if nonbasic and strictly between its bounds, BS if basic and SBS if superbasic).
A key is sometimes printed before State.
Note that unless the optional parameter Scale Option = 0${\mathbf{Scale\; Option}}=0$ is specified, the tests for assigning a key are applied to the variables of the scaled problem.
A 
Alternative optimum possible. The variable is nonbasic, but its reduced gradient is essentially zero. This means that if the variable were allowed to start moving away from its bound, there would be no change in the value of the objective function. The values of the other free variables might change, giving a genuine alternative solution. However, if there are any degenerate variables (labelled D), the actual change might prove to be zero, since one of them could encounter a bound immediately. In either case, the values of the Lagrange multipliers might also change.

D 
Degenerate. The variable is basic or superbasic, but it is equal (or very close) to one of its bounds.

I 
Infeasible. The variable is basic or superbasic and is currently violating one of its bounds by more than the value of the Feasibility Tolerance.

N 
Not precisely optimal. The variable is nonbasic or superbasic. If the value of the reduced gradient for the variable exceeds the value of the optional parameter Optimality Tolerance, the solution would not be declared optimal because the reduced gradient for the variable would not be considered negligible.


Activity 
is the value of x_{j}${x}_{j}$ at the final iterate.

Obj Gradient 
is the value of g_{j}${g}_{j}$ at the final iterate. For FP problems, g_{j}${g}_{j}$ is set to zero.

Lower Bound 
is the lower bound specified for the variable. None indicates that bl(j) ≤ − bigbnd${\mathbf{bl}}\left(j\right)\le \mathit{bigbnd}$.

Upper Bound 
is the upper bound specified for the variable. None indicates that bu(j) ≥ bigbnd${\mathbf{bu}}\left(j\right)\ge \mathit{bigbnd}$.

Reduced Gradnt 
is the value of d_{j}${d}_{j}$ at the final iterate (see Section [Main Iteration]). For FP problems, d_{j}${d}_{j}$ is set to zero.

m + j 
is the value of m + j$m+j$.

Let
v_{i}${v}_{\mathit{i}}$ denote the
i$\mathit{i}$th row of
A$A$, for
i = 1,2, … ,m$\mathit{i}=1,2,\dots ,m$. The following describes the printout for each row (or constraint). A full stop (.) is printed for any numerical value that is zero.
Number 
is the value of n + i$n+i$. (This is used internally to refer to s_{i}${s}_{i}$ in the intermediate output.)

Row 
gives the name of ν_{i}${\nu}_{i}$.

State 
gives the state of v_{i}${v}_{i}$ (LL if active on its lower bound, UL if active on its upper bound, EQ if active and fixed, BS if inactive when s_{i}${s}_{i}$ is basic and SBS if inactive when s_{i}${s}_{i}$ is superbasic).
A key is sometimes printed before State.
Note that unless the optional parameter Scale Option = 0${\mathbf{Scale\; Option}}=0$ is specified, the tests for assigning a key are applied to the variables of the scaled problem.
A 
Alternative optimum possible. The variable is nonbasic, but its reduced gradient is essentially zero. This means that if the variable were allowed to start moving away from its bound, there would be no change in the value of the objective function. The values of the other free variables might change, giving a genuine alternative solution. However, if there are any degenerate variables (labelled D), the actual change might prove to be zero, since one of them could encounter a bound immediately. In either case, the values of the Lagrange multipliers might also change.

D 
Degenerate. The variable is basic or superbasic, but it is equal (or very close) to one of its bounds.

I 
Infeasible. The variable is basic or superbasic and is currently violating one of its bounds by more than the value of the Feasibility Tolerance.

N 
Not precisely optimal. The variable is nonbasic or superbasic. If the value of the reduced gradient for the variable exceeds the value of the optional parameter Optimality Tolerance, the solution would not be declared optimal because the reduced gradient for the variable would not be considered negligible.


Activity 
is the value of v_{i}${v}_{i}$ at the final iterate.

Slack Activity 
is the value by which the row differs from its nearest bound. (For the free row (if any), it is set to Activity.)

Lower Bound 
is the lower bound specified for the variable. None indicates that bl(j) ≤ − bigbnd${\mathbf{bl}}\left(j\right)\le \mathit{bigbnd}$.

Upper Bound 
is the upper bound specified for the variable. None indicates that bu(j) ≥ bigbnd${\mathbf{bu}}\left(j\right)\ge \mathit{bigbnd}$.

Dual Activity 
is the value of the dual variable π_{i}${\pi}_{i}$ (the Lagrange multiplier for ν_{i}${\nu}_{i}$; see Section [Main Iteration]). For FP problems, π_{i}${\pi}_{i}$ is set to zero.

i 
gives the index i$i$ of the i$i$th row.

Numerical values are output with a fixed number of digits; they are not guaranteed to be accurate to this precision.
PDF version (NAG web site
, 64bit version, 64bit version)
© The Numerical Algorithms Group Ltd, Oxford, UK. 2009–2013