Before calling nag_opt_qpconvex1_sparse_solve (e04nk), or the option setting function nag_opt_qpconvex1_sparse_option_string (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:

\underset{x \in R^{n}}{minimize} f (x) subject to l \leq \{\begin{matrix} x \\ A x \end{matrix}\} \leq u,

(1)

where

x

is a set of variables,

A

is an

m

n

matrix and the objective function

f (x)

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)

is maximized. The possible forms for

f (x)

are listed in Table 1, in which the prefixes FP, LP and QP stand for ‘feasible point’, ‘linear programming’ and ‘quadratic programming’ respectively,

c

is an

n

-element vector and

H

is the

n

n

second-derivative matrix

\nabla^{2} f (x)

(the Hessian matrix).

Problem type	Objective function $f (x)$	Hessian matrix $H$
FP	Not applicable	Not applicable
LP	$c^{T} x$	Not applicable
QP	$c^{T} x + \frac{1}{2} x^{T} H x$	Symmetric positive semidefinite

Table 1

For LP and QP problems, the unique global minimum value of

f (x)

is found. For FP problems,

f (x)

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

H x

for any given vector

x

. (

H

need not be stored explicitly.) If

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

(see Arguments).

The defining feature of a convex QP problem is that the matrix

H

must be positive semidefinite, i.e., it must satisfy

x^{T} H x \geq 0

for all

x

. Otherwise,

f (x)

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 large-scale linear and quadratic programming problems in which the constraint matrix

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

. (Sparsity in

H

can be exploited in the function that computes

H x

.) For problems in which

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

elements of

A x

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

, where

s = {(s_{1}, s_{2}, \dots, s_{m})}^{T}

. For example, the linear constraint

5 \leq 2 x_{1} + 3 x_{2} \leq + \infty

is replaced by

2 x_{1} + 3 x_{2} - s_{1} = 0

, together with the bounded slack

5 \leq s_{1} \leq + \infty

. The problem defined by (1) can therefore be re-written in the following equivalent form:

\underset{x \in R^{n}, s \in R^{m}}{minimize} f (x) subject to A x - s = 0, l \leq \{\begin{matrix} x \\ s \end{matrix}\} \leq u .

Since the slack variables

s

are subject to the same upper and lower bounds as the elements of

A x

, the bounds on

A x

and

x

can simply be thought of as bounds on the combined vector

(x, s)

. (In order to indicate their special role in QP problems, the original variables

x

are sometimes known as ‘column variables’, and the slack variables

s

are known as ‘row variables’.)

Each LP or QP problem is solved using an active-set 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)

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

if the associated element of either

x

A x

is equal to one of its upper or lower bounds. Since an active constraint in

A x

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)

. 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 active-set method, the constraints

A x - s = 0

are (conceptually) partitioned into the form

B x_{B} + S x_{S} + N x_{N} = 0,

where

x_{N}

consists of the nonbasic elements of

(x, s)

and the basis matrix

B

is square and nonsingular. The elements of

x_{B}

and

x_{S}

are called the basic and superbasic variables respectively; with

x_{N}

they are a permutation of the elements of

x

and

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}

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)

continues to satisfy

A x - s = 0

. The number of superbasic variables (

n_{S}

say) therefore indicates the number of degrees of freedom remaining after the constraints have been satisfied. In broad terms,

n_{S}

is a measure of how nonlinear the problem is. In particular,

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

S

and

N

, a nonbasic variable is selected to be added to

S

, and the process is repeated with the value of

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}

is decreased by one.

Associated with each of the

m

equality constraints

A x - s = 0

is a dual variable

π_{i}

. Similarly, each variable in

(x, s)

has an associated reduced gradient

d_{j}

(also known as a reduced cost). The reduced gradients for the variables

x

are the quantities

g - A^{T} π

, where

g

is the gradient of the QP objective function; and the reduced gradients for the slack variables

s

are the dual variables

π

. The QP subproblem is optimal if

d_{j} \geq 0

for all nonbasic variables at their lower bounds,

d_{j} \leq 0

for all nonbasic variables at their upper bounds and

d_{j} = 0

for all superbasic variables. In practice, an approximate QP solution is found by slightly relaxing these conditions on

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}

means that the reduced gradient

d_{j}

associated with the relevant active upper or lower bound on

x_{j}

is computed via the formula

d_{j} = g_{j} - a^{T} π

, where

a_{j}

is the

j

th column of

(\begin{matrix} A & - I \end{matrix})

. (The variable selected by such a process and the corresponding value of

d_{j}

(i.e., its reduced gradient) are the quantities +S and dj in the monitoring file output; see Description of Monitoring Information.) If

A

has significantly more columns than rows (i.e.,

n ≫ 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}

'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

L U

factors of the basis matrix

B

), a practical anti-degeneracy procedure, efficient handling of linear constraints and bounds on the variables (by an active-set strategy), as well as automatic scaling of the constraints. Further details can be found in 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 Large-scale 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 anti-cycling procedure for linearly constrained optimization Math. Programming 45 437–474

Gill P E, Murray W, Saunders M A and Wright M H (1991) Inertia-controlling 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 83-20R Department of Operations Research, Stanford University

Parameters

Compulsory Input Parameters

1: $n$ – int64int32nag_int scalar

n

, the number of variables (excluding slacks). This is the number of columns in the linear constraint matrix

A

Constraint:

n \geq 1

2: $m$ – int64int32nag_int scalar

m

, the number of general linear constraints (or slacks). This is the number of rows in

A

, including the free row (if any; see iobj).

Constraint:

m \geq 1

3: $iobj$ – int64int32nag_int scalar

iobj > 0

, row iobj of

A

is a free row containing the nonzero elements of the vector

c

appearing in the linear objective term

c^{T} x

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

Constraint:

0 \leq iobj \leq m

4: $ncolh$ – int64int32nag_int scalar

n_{H}

, the number of leading nonzero columns of the Hessian matrix

H

. For FP and LP problems, ncolh must be set to zero.

Constraint:

0 \leq ncolh \leq n

5: $qphx$ – function handle or string containing name of m-file

For QP problems, you must supply a version of qphx to compute the matrix product

H x

. If

H

has zero rows and columns, it is most efficient to order the variables

x = {(\begin{matrix} y & z \end{matrix})}^{T}

so that

H x = (\begin{matrix} H_{1} & 0 \\ 0 & 0 \end{matrix}) (\begin{matrix} y \\ z \end{matrix}) = (\begin{matrix} H_{1} y \\ 0 \end{matrix}),

where the nonlinear variables

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 nag_opt_qpconvex1_sparse_dummy_qphx_old (e04nku).

[hx, user] = qphx(nstate, ncolh, x, user)

Input Parameters

1: $nstate$ – int64int32nag_int scalar: If $nstate = 1$ , nag_opt_qpconvex1_sparse_solve (e04nk) is calling qphx for the first time. This argument setting allows you to save computation time if certain data must be read or calculated only once.
If $nstate \geq 2$ , nag_opt_qpconvex1_sparse_solve (e04nk) is calling qphx for the last time. This argument setting allows you to perform some additional computation on the final solution. In general, the last call to qphx is made with $nstate = 2 + ifail$ (see Error Indicators and Warnings).

Otherwise, $nstate = 0$ .
2: $ncolh$ – int64int32nag_int scalar: This is the same argument ncolh as supplied to nag_opt_qpconvex1_sparse_solve (e04nk).
3: $x (ncolh)$ – double array: The first ncolh elements of the vector $x$ .
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 $H x$ .
2: $user$ – Any MATLAB object

6: $a (nz)$ – double array

The nonzero elements of

A

, ordered by increasing column index. Note that elements with the same row and column indices are not allowed.

7: $ha (nz)$ – int64int32nag_int array

ha (i)

must contain the row index of the nonzero element stored in

a (i)

, for

i = 1, 2, \dots, nz

. Note that the row indices for a column may be supplied in any order.

Constraint:

1 \leq ha (i) \leq m

, for

i = 1, 2, \dots, nz

8: $ka (n + 1)$ – int64int32nag_int array

ka (j)

must contain the index in a of the start of the

j

th column, for

j = 1, 2, \dots, n

ka (n + 1)

must be set to

nz + 1

. To specify the

j

th column as empty, set

ka (j) = ka (j + 1)

. As a consequence

ka (1)

is always

1

Constraints:

$ka (1) = 1$ ;
$ka (j) \geq 1$ , for $j = 2, 3, \dots, n$ ;
$ka (n + 1) = nz + 1$ ;
$0 \leq ka (j + 1) - ka (j) \leq m$ , for $j = 1, 2, \dots, n$ .

9: $bl (n + m)$ – double array

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

, and the next m elements the bounds for the general linear constraints

A x

(or slacks

s

) and the free row (if any). To specify a nonexistent lower bound (i.e.,

l_{j} = - \infty

), set

bl (j) \leq - bigbnd

, where

bigbnd

is the value of the optional parameter Infinite Bound Size. To specify the

j

th constraint as an equality, set

bl (j) = bu (j) = β

, say, where

|β| < bigbnd

. Note that the lower bound corresponding to the free row must be set to

- \infty

and stored in

bl (n + iobj)

Constraint: if

iobj > 0

bl (n + iobj) \leq - bigbnd

(See also the description for bu.)

10: $bu (n + m)$ – double array

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

, and the next m elements the bounds for the general linear constraints

A x

(or slacks

s

) and the free row (if any). To specify a nonexistent upper bound (i.e.,

u_{j} = + \infty

), set

bu (j) \geq bigbnd

. Note that the upper bound corresponding to the free row must be set to

+ \infty

and stored in

bu (n + iobj)

Constraints:

if $iobj > 0$ , $bu (n + iobj) \geq bigbnd$ ;
$bl (j) \leq bu (j)$ , for $j = 1, 2, \dots, n + m$ ;
if $bl (j) = bu (j) = β$ , $|β| < bigbnd$ .

11: $start$ – string (length ≥ 1)

Indicates how a starting basis is to be obtained.

$start ='C'$: An internal Crash procedure will be used to choose an initial basis matrix $B$ .
$start ='W'$: A basis is already defined in istate (probably from a previous call).

Constraint:

start ='C'

'W'

12: $names (5)$ – cell array of strings

A set of names associated with the so-called MPSX form of the problem, as follows:

$names (1)$: Must contain the name for the problem (or be blank).
$names (2)$: Must contain the name for the free row (or be blank).
$names (3)$: Must contain the name for the constraint right-hand side (or be blank).
$names (4)$: Must contain the name for the ranges (or be blank).
$names (5)$: Must contain the name for the bounds (or be blank).

(These names are used in the monitoring file output; see Description of Monitoring Information.)

13: $crname (nname)$ – cell array of strings

The optional column and row names, respectively.

nname = 1

, crname is not referenced and the printed output will use default names for the columns and rows.

nname = n + 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)

14: $ns$ – int64int32nag_int scalar

n_{S}

, the number of superbasics. For QP problems, ns need not be specified if

start ='C'

, but must retain its value from a previous call when

start ='W'

. For FP and LP problems, ns need not be initialized.

15: $xs (n + m)$ – double array

The initial values of the variables and slacks

(x, s)

. (See the description for istate.)

16: $istate (n + m)$ – int64int32nag_int array

start ='C'

, the first n elements of istate and xs must specify the initial states and values, respectively, of the variables

x

. (The slacks

s

need not be initialized.) An internal Crash procedure is then used to select an initial basis matrix

B

. The initial basis matrix will be triangular (neglecting certain small elements in each column). It is chosen from various rows and columns of

(\begin{matrix} A & - I \end{matrix})

. Possible values for

istate (j)

are as follows:

$istate (j)$	State of $xs (j)$ during Crash procedure
$0$ or $1$	Eligible for the basis
$2$	Ignored
$3$	Eligible for the basis (given preference over $0$ or $1$ )
$4$ or $5$	Ignored

If nothing special is known about the problem, or there is no wish to provide special information, you may set

istate (j) = 0

and

xs (j) = 0.0

, for

j = 1, 2, \dots, n

. All variables will then be eligible for the initial basis. Less trivially, to say that the

j

th variable will probably be equal to one of its bounds, set

istate (j) = 4

and

xs (j) = bl (j)

istate (j) = 5

and

xs (j) = bu (j)

as appropriate.

Following the Crash procedure, variables for which

istate (j) = 2

are made superbasic. Other variables not selected for the basis are then made nonbasic at the value

xs (j)

bl (j) \leq xs (j) \leq bu (j)

, or at the value

bl (j)

bu (j)

closest to

xs (j)

start ='W'

, istate and xs must specify the initial states and values, respectively, of the variables and slacks

(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 $start ='C'$ , $0 \leq istate (j) \leq 5$ , for $j = 1, 2, \dots, n$ ;
if $start ='W'$ , $0 \leq istate (j) \leq 3$ , for $j = 1, 2, \dots, n + m$ .

17: $leniz$ – int64int32nag_int scalar

The dimension of the array iz.

Constraint:

leniz \geq 1

18: $lenz$ – int64int32nag_int scalar

The dimension of the array z.

Constraint:

lenz \geq 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

. (nag_opt_qpconvex1_sparse_solve (e04nk) will then terminate with

ifail = 12

19: $lwsav (20)$ – logical array

20: $iwsav (380)$ – int64int32nag_int array

21: $rwsav (285)$ – 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), nag_opt_qpconvex1_sparse_option_string (e04nm) or nag_opt_init (e04wb).

Optional Input Parameters

1: $nz$ – 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

Constraint:

1 \leq nz \leq n \times 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.

$nname = 1$: There are no names. Default names will be used in the printed output.
$nname = n + m$: All names must be supplied.

Constraint:

nname = 1

n + 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 m-files than to use user.

Output Parameters

1: $ns$ – int64int32nag_int scalar

The final number of superbasics. This will be zero for FP and LP problems.

2: $xs (n + m)$ – double array

The final values of the variables and slacks

(x, s)

3: $istate (n + m)$ – int64int32nag_int array

The final states of the variables and slacks

(x, s)

. The significance of each possible value of

istate (j)

is as follows:

$istate (j)$	State of variable $j$	Normal value of $xs (j)$
$0$	Nonbasic	$bl (j)$
$1$	Nonbasic	$bu (j)$
$2$	Superbasic	Between $bl (j)$ and $bu (j)$
$3$	Basic	Between $bl (j)$ and $bu (j)$

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

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

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)

lies strictly between its bounds.

ninf > 0

, some basic and superbasic variables may be outside their bounds by an arbitrary amount (bounded by sinf if

Scale Option = 0

4: $miniz$ – int64int32nag_int scalar

The minimum value of leniz required to start solving the problem. If

ifail = 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

, 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

1

7: $sinf$ – double scalar

The sum of infeasibilities. This will be zero if

ninf = 0

. (Note that nag_opt_qpconvex1_sparse_solve (e04nk) does not attempt to compute the minimum value of sinf if

ifail = 3

8: $obj$ – double scalar

The value of the objective function.

ninf = 0

, obj includes the quadratic objective term

\frac{1}{2} x^{T} H x

(if any).

ninf > 0

, obj is just the linear objective term

c^{T} x

(if any).

For FP problems, obj is set to zero.

9: $clamda (n + m)$ – 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: $lwsav (20)$ – logical array

12: $iwsav (380)$ – int64int32nag_int array

13: $rwsav (285)$ – double array

14: $ifail$ – int64int32nag_int scalar

ifail = 0

unless the function detects an error (see Error Indicators and Warnings).

nag_opt_qpconvex1_sparse_solve (e04nk) returns with

ifail = 0

if the reduced gradient (Norm rg; see Printed output) is negligible, the Lagrange multipliers (Lagr Mult; see Printed output) are optimal and

x

satisfies the constraints to the accuracy requested by the value of the optional parameter Feasibility Tolerance (

default value = \max (10^{- 6}, \sqrt{ε})

, where

ε

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 $ifail = 1$: Weak solution found. The final $x$ is not unique, although $x$ gives the global minimum value of the objective function.

W $ifail = 2$: The problem is unbounded (or badly scaled). The objective function is not bounded below in the feasible region.

W $ifail = 3$: The problem is infeasible. The general constraints cannot all be satisfied simultaneously to within the value of the optional parameter Feasibility Tolerance ( $default value = \max (10^{- 6}, \sqrt{ε})$ , where $ε$ is the machine precision).

$ifail = 4$: Too many iterations. The value of the optional parameter Iteration Limit ( $default value = \max (50, 5 (n + m))$ ) is too small.

$ifail = 5$: The reduced Hessian matrix $Z^{T} H Z$ (see Definition of the Working Set and Search Direction) exceeds its assigned dimension. The value of the optional parameter Superbasics Limit ( $default value = \min (n_{H} + 1, n)$ ) is too small.

$ifail = 6$: The Hessian matrix $H$ appears to be indefinite. This sometimes occurs because the values of the optional parameters LU Factor Tolerance ( $default value = 100.0$ ) and LU Update Tolerance ( $default value = 10.0$ ) are too large. Check also that qphx has been coded correctly and that all relevant elements of $H x$ have been assigned their correct values.

$ifail = 7$: An input argument is invalid.

$ifail = 8$: Numerical error in trying to satisfy the general constraints. The basis is very ill-conditioned.

$ifail = 9$: Not enough integer workspace for the basis factors. Increase leniz and rerun nag_opt_qpconvex1_sparse_solve (e04nk).

$ifail = 10$: Not enough real workspace for the basis factors. Increase lenz and rerun nag_opt_qpconvex1_sparse_solve (e04nk).

$ifail = 11$: The basis is singular after $15$ attempts to factorize it (adding slacks where necessary). Either the problem is badly scaled or the value of the optional parameter LU Factor Tolerance ( $default value = 100.0$ ) is too large.

$ifail = 12$: Not enough integer workspace to start solving the problem. Increase leniz to at least miniz and rerun nag_opt_qpconvex1_sparse_solve (e04nk).

$ifail = 13$: Not enough real workspace to start solving the problem. Increase lenz to at least minz and rerun nag_opt_qpconvex1_sparse_solve (e04nk).

$ifail = - 99$: An unexpected error has been triggered by this routine. Please contact NAG.

$ifail = - 399$: Your licence key may have expired or may not have been installed correctly.

$ifail = - 999$: Dynamic memory allocation failed.

Accuracy

nag_opt_qpconvex1_sparse_solve (e04nk) implements a numerically stable active-set 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

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 $x$ is not feasible, Sinf gives the sum of the magnitudes of constraint violations. If $x$ 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 $‖d_{S}‖$ , the Euclidean norm of the reduced gradient (see 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

, a default name is assigned to the

j

th variable, for

j = 1, 2, \dots, n

. If

nname = n + m

, the name supplied in

crname (j)

is assigned to the

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

(

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) \leq - bigbnd

Upper Bound

is the upper bound specified for the variable. None indicates that

bu (j) \geq bigbnd

Lagr Mult

is the Lagrange multiplier for the associated bound. This will be zero if State is FR. If

x

is optimal, the multiplier should be non-negative if State is LL, non-positive 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)

and

bu (j)

. A blank entry indicates that the associated variable is not bounded (i.e.,

bl (j) \leq - bigbnd

and

bu (j) \geq bigbnd

The meaning of the printout for linear constraints is the same as that given above for variables, with ‘variable’ replaced by ‘constraint’,

n

replaced by

m

crname (j)

replaced by

crname (n + j)

bl (j)

and

bu (j)

are replaced by

bl (n + j)

and

bu (n + j)

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

This example minimizes the quadratic function

f (x) = c^{T} x + \frac{1}{2} x^{T} H x

, where

c = {(- 200.0, - 2000.0, - 2000.0, - 2000.0, - 2000.0, 400.0, 400.0)}^{T}

H = (\begin{matrix} 2 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 2 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 2 & 2 & 0 & 0 & 0 \\ 0 & 0 & 2 & 2 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 2 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 2 & 2 \\ 0 & 0 & 0 & 0 & 0 & 2 & 2 \end{matrix})

subject to the bounds

\begin{matrix} 0 \leq x_{1} \leq 200 \\ 0 \leq x_{2} \leq 2500 \\ 400 \leq x_{3} \leq 800 \\ 100 \leq x_{4} \leq 700 \\ 0 \leq x_{5} \leq 1500 \\ 0 \leq x_{6} \\ 0 \leq x_{7} \end{matrix}

and to the linear constraints

\begin{matrix} x_{1} + x_{2} + x_{3} + x_{4} + x_{5} + x_{6} + x_{7} = 2000 \\ 0.15 x_{1} + 0.04 x_{2} + 0.02 x_{3} + 0.04 x_{4} + 0.02 x_{5} + 0.01 x_{6} + 0.03 x_{7} \leq 60 \\ 0.03 x_{1} + 0.05 x_{2} + 0.08 x_{3} + 0.02 x_{4} + 0.06 x_{5} + 0.01 x_{6} \leq 100 \\ 0.02 x_{1} + 0.04 x_{2} + 0.01 x_{3} + 0.02 x_{4} + 0.02 x_{5} \leq 40 \\ 0.02 x_{1} + 0.03 x_{2} + 0.01 x_{5} \leq 30 \\ 1500 \leq 0.70 x_{1} + 0.75 x_{2} + 0.80 x_{3} + 0.75 x_{4} + 0.80 x_{5} + 0.97 x_{6} \\ 250 \leq 0.02 x_{1} + 0.06 x_{2} + 0.08 x_{3} + 0.12 x_{4} + 0.02 x_{5} + 0.01 x_{6} + 0.97 x_{7} \leq 300 \end{matrix}

The initial point, which is infeasible, is

x_{0} = {(0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0)}^{T} .

The optimal solution (to five figures) is

x^{*} = {(0.0, 349.40, 648.85, 172.85, 407.52, 271.36, 150.02)}^{T} .

One bound constraint and four linear constraints are active at the solution. Note that the Hessian matrix

H

is positive semidefinite.

Open in the MATLAB editor: e04nk_example

function e04nk_example


fprintf('e04nk example results\n\n');

n = int64(7);
m = int64(8);
iobj = int64(8);
ncolh = int64(7);

% A in coordinate storage form to see the problem better
a = [ 1    1    1    1    1    1    1    ...
      0.15 0.04 0.02 0.04 0.02 0.01 0.03 ...
      0.03 0.05 0.08 0.02 0.06 0.01      ...
      0.02 0.04 0.01 0.02 0.02           ...
      0.02 0.03           0.01           ...
      0.70 0.75 0.80 0.75 0.80 0.97      ...
      0.02 0.06 0.08 0.12 0.02 0.01 0.97 ...
     -200 -2000 -2000 -2000 -2000 400 400];
irow = int64([1  1  1  1  1  1  1 ...
                2  2  2  2  2  2  2 ...
                3  3  3  3  3  3    ...
                4  4  4  4  4       ...
                5  5        5       ...
                6  6  6  6  6  6    ...
                7  7  7  7  7  7  7 ...
                8  8  8  8  8  8  8]);
icol = int64([1  2  3  4  5  6  7 ...
                1  2  3  4  5  6  7 ...
                1  2  3  4  5  6    ...
                1  2  3  4  5       ...
                1  2        5       ...
                1  2  3  4  5  6    ...
                1  2  3  4  5  6  7 ...
                1  2  3  4  5  6  7]);

% Convert to compressed column storage format for e04nk
dup = 'F';
zero = 'R';
nz = int64(size(a,2));
[nz, a, irow, ha, ka, ifail] = ...
f11za( ...
       m, nz, a, icol, irow, dup, zero);

% constraint bounds
bn = -1e25; bp = 1e25;
bl = [  0    0 400 100    0  0  0 2000 bn  bn bn bn 1500 250 bn];
bu = [200 2500 800 700 1500 bp bp 2000 60 100 40 30   bp 300 bp];
start = 'C';
names = {'        '; '        '; '        '; '        '; '        '};
crname = {'...X1...'; '...X2...'; '...X3...'; '...X4...'; '...X5...'; ...
          '...X6...'; '...X7...'; '..ROW1..'; '..ROW2..'; '..ROW3..'; ...
          '..ROW4..'; '..ROW5..'; '..ROW6..'; '..ROW7..'; '..COST..'};


ns = int64(0);
xs = zeros(n+m,1);
istate = zeros(n+m, 1, 'int64');
leniz = int64(10000);
lenz  = int64(10000);

% Initialize
[cwsav,lwsav,iwsav,rwsav,ifail] = e04wb( ...
                                         'e04nk');
% Print solution option
[lwsav, iwsav, rwsav, inform] = e04nm( ...
                                       'Print Level = 1', lwsav, iwsav, rwsav);
% Optimize
[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);



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);

e04nk example results


 *** E04NKA

 Parameters
 ----------

 Frequencies.
 Check frequency.........        60       Expand frequency........     10000
 Factorization frequency.       100

 LP Parameters.
 Scale tolerance.........  9.00E-01       Feasibility tolerance...  1.00E-06
 Iteration limit.........        75       Scale option............         2
 Optimality tolerance....  1.00E-06       Partial price...........        10
 Crash tolerance.........  1.00E-01       Pivot tolerance.........  2.04E-11
 Crash option............         2

 QP objective.
 Objective variables.....         7       Hessian columns.........         7
 Superbasics limit.......         7

 Miscellaneous.
 Variables...............         7       Linear constraints......         8
 LU factor tolerance.....  1.00E+02       LU update tolerance.....  1.00E+01
 LU singularity tolerance  2.04E-11       Monitoring file.........        -1
 EPS (machine precision).  1.11E-16       Print level.............         1
 Infinite bound size.....  1.00E+20       Infinite step size......  1.00E+20
 COLD start..............                 MINIMIZE................

 Workspace provided is                 IZ(   10000),  Z(   10000).
 To start solving the problem we need  IZ(     428),  Z(     358).


 Variable State     Value       Lower Bound  Upper Bound   Lagr Mult  Residual

 ...X1...    LL    0.00000           .        200.00       2361.         .
 ...X2...    BS    349.399           .        2500.0     -3.4207E-12   349.4
 ...X3...   SBS    648.853       400.00       800.00     -2.2932E-12   151.1
 ...X4...   SBS    172.847       100.00       700.00      1.9491E-12   72.85
 ...X5...    BS    407.521           .        1500.0      9.1881E-13   407.5
 ...X6...    BS    271.356           .           None    -1.4910E-13   271.4
 ...X7...    BS    150.023           .           None     1.5700E-12   150.0


 Constrnt State     Value       Lower Bound  Upper Bound   Lagr Mult  Residual

 ..ROW1..    EQ    2000.00       2000.0       2000.0     -1.2901E+04     .
 ..ROW2..    BS    49.2316          None      60.000           .      -10.77
 ..ROW3..    UL    100.000          None      100.00      -2325.         .
 ..ROW4..    BS    32.0719          None      40.000           .      -7.928
 ..ROW5..    BS    14.5572          None      30.000           .      -15.44
 ..ROW6..    LL    1500.00       1500.0          None     1.4455E+04     .
 ..ROW7..    LL    250.000       250.00       300.00      1.4581E+04     .
 ..COST..    BS  -2.988690E+06      None         None     -1.000     -2.9887E+06

 Exit E04NKA - Optimal QP solution found.

 Final QP objective value =    -1847785.

 Exit from QP problem after       9 iterations.

Note: the remainder of this document is intended for more advanced users. Algorithmic Details contains a detailed description of the algorithm which may be needed in order to understand Optional Parameters and Description of Monitoring Information. Optional Parameters describes the optional parameters which may be set by calls to nag_opt_qpconvex1_sparse_option_string (e04nm). 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 inertia-controlling 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 arguments 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 two-phase nature of the algorithm is reflected by changing the function being minimized from the sum of infeasibilities (the printed quantity Sinf; see Description of Monitoring Information) to the quadratic objective function (the printed quantity Objective; see Description of Monitoring Information).

In general, an iterative process is required to solve a quadratic program. Given an iterate

(x, s)

in both the original variables

x

and the slack variables

s

, a new iterate

(\bar{x}, \bar{s})

is defined by

(\begin{matrix} \bar{x} \\ \bar{s} \end{matrix}) = (\begin{matrix} x \\ s \end{matrix}) + α p,

(2)

where the step length

α

is a non-negative scalar (the printed quantity Step; see Description of Monitoring Information), and

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)

, 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}

denote the number of constraints in the working set (including bounds), and let

W

denote the associated

m_{W}

(n + m)

working set matrix consisting of the

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

must satisfy the identity

W p = 0 .

(3)

This characterisation allows

p

to be computed using any

n

n_{Z}

full-rank matrix

Z

that spans the null space of

W

. (Thus,

n_{Z} = n - m_{W}

and

W Z = 0

.) The null space matrix

Z

is defined from a sparse

L U

factorization of part of

W

(see (6) and (7)). The direction

p

will satisfy (3) if

p = Z p_{Z},

(4)

where

p_{Z}

is any

n_{Z}

-vector.

The working set contains the constraints

A x - s = 0

and a subset of the upper and lower bounds on the variables

(x, s)

. Since the gradient of a bound constraint

x_{j} \geq l_{j}

x_{j} \leq u_{j}

is a vector of all zeros except for

\pm 1

in position

j

, it follows that the working set matrix contains the rows of

(\begin{matrix} A & - I \end{matrix})

and the unit rows associated with the upper and lower bounds in the working set.

The working set matrix

W

can be represented in terms of a certain column partition of the matrix

(\begin{matrix} A & - I \end{matrix})

by (conceptually) partitioning the constraints

A x - s = 0

so that

B x_{B} + S x_{S} + N x_{N} = 0,

(5)

where

B

is a square nonsingular basis and

x_{B}

x_{S}

and

x_{N}

are the basic, superbasic and nonbasic variables respectively. The nonbasic variables are equal to their upper or lower bounds at

(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}

(the printed quantity Ns; see Description of Monitoring Information). Given values of

x_{N}

and

x_{S}

, the basic variables

x_{B}

are adjusted so that

(x, s)

satisfies (5).

P

is a permutation matrix such that

(\begin{matrix} A & - I \end{matrix}) P = (\begin{matrix} B & S & N \end{matrix})

, then

W

satisfies

W P = (\begin{matrix} B & S & N \\ 0 & 0 & I_{N} \end{matrix}),

(6)

where

I_{N}

is the identity matrix with the same number of columns as

N

The null space matrix

Z

is defined from a sparse

L U

factorization of part of

W

. In particular,

Z

is maintained in ‘reduced gradient’ form, using the LUSOL package (see Gill et al. (1991)) to maintain sparse

L U

factors of the basis matrix

B

that alters as the working set

W

changes. Given the permutation

P

, the null space basis is given by

Z = P (\begin{matrix} - B^{- 1} S \\ I \\ 0 \end{matrix}) .

(7)

This matrix is used only as an operator, i.e., it is never computed explicitly. Products of the form

Z v

and

Z^{T} g

are obtained by solving with

B

B^{T}

. This choice of

Z

implies that

n_{Z}

, the number of ‘degrees of freedom’ at

(x, s)

, is the same as

n_{S}

, the number of superbasic variables.

Let

g_{Z}

and

H_{Z}

denote the reduced gradient and reduced Hessian of the objective function:

g_{Z} = Z^{T} g and H_{Z} = Z^{T} H Z,

(8)

where

g

is the objective gradient at

(x, s)

. Roughly speaking,

g_{Z}

and

H_{Z}

describe the first and second derivatives of an

n_{S}

-dimensional unconstrained problem for the calculation of

p_{Z}

. (The condition estimator of

H_{Z}

is the quantity Cond Hz in the monitoring file output; see Description of Monitoring Information.)

At each iteration, an upper triangular factor

R

is available such that

H_{Z} = R^{T} R

. Normally,

R

is computed from

R^{T} R = Z^{T} H Z

at the start of the optimality phase and then updated as the QP working set changes. For efficiency, the dimension of

R

should not be excessive (say,

n_{S} \leq 1000

). This is guaranteed if the number of nonlinear variables is ‘moderate’.

If the QP problem contains linear variables,

H

is positive semidefinite and

R

may be singular with at least one zero diagonal element. However, an inertia-controlling strategy is used to ensure that only the last diagonal element of

R

can be zero. (See Gill et al. (1991) for a discussion of a similar strategy for indefinite quadratic programming.)

If the initial

R

is singular, enough variables are fixed at their current value to give a nonsingular

R

. This is equivalent to including temporary bound constraints in the working set. Thereafter,

R

can become singular only when a constraint is deleted from the working set (in which case no further constraints are deleted until

R

becomes nonsingular).

Main Iteration

If the reduced gradient is zero,

(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

minimizes the quadratic objective function when the constraints in the working set are treated as equalities. At a constrained stationary point, Lagrange multipliers

λ

are defined from the equations

W^{T} λ = g (x) .

(9)

A Lagrange multiplier

λ_{j}

corresponding to an inequality constraint in the working set is said to be optimal if

λ_{j} \leq σ

when the associated constraint is at its upper bound, or if

λ_{j} \geq - σ

when the associated constraint is at its lower bound, where

σ

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

(see Error Indicators and Warnings).

The special form (6) of the working set allows the multiplier vector

λ

, the solution of (9), to be written in terms of the vector

d = (\begin{matrix} g \\ 0 \end{matrix}) - {(\begin{matrix} A & - I \end{matrix})}^{T} π = (\begin{matrix} g - A^{T} π \\ π \end{matrix}),

(10)

where

π

satisfies the equations

B^{T} π = g_{B}

, and

g_{B}

denotes the basic elements of

g

. The elements of

π

are the Lagrange multipliers

λ_{j}

associated with the equality constraints

A x - s = 0

. The vector

d_{N}

of nonbasic elements of

d

consists of the Lagrange multipliers

λ_{j}

associated with the upper and lower bound constraints in the working set. The vector

d_{S}

of superbasic elements of

d

is the reduced gradient

g_{Z}

in (8). The vector

d_{B}

of basic elements of

d

is zero, by construction. (The Euclidean norm of

d_{S}

and the final values of

d_{S}

g

and

π

are the quantities Norm rg, Reduced Gradnt, Obj Gradient and Dual Activity in the monitoring file output; see 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 = 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}

, depending on whether or not

H_{Z}

is singular. If

H_{Z}

is nonsingular,

R

is nonsingular and

p_{Z}

in (4) is computed from the equations

R^{T} R p_{Z} = - g_{Z},

(11)

where

g_{Z}

is the reduced gradient at

x

. In this case,

(x, s) + p

is the minimizer of the objective function subject to the working set constraints being treated as equalities. If

(x, s) + p

is feasible,

α

is defined to be unity. In this case, the reduced gradient at

(\bar{x}, \bar{s})

will be zero, and Lagrange multipliers are computed at the next iteration. Otherwise,

α

is set to

α_{m}

, the step to the ‘nearest’ constraint along

p

. This constraint is then added to the working set at the next iteration.

H_{Z}

is singular, then

R

must also be singular, and an inertia-controlling strategy is used to ensure that only the last diagonal element of

R

is zero. (See Gill et al. (1991) for a discussion of a similar strategy for indefinite quadratic programming.) In this case,

p_{Z}

satisfies

p_{Z}^{T} H_{Z} p_{Z} = 0 and g_{Z}^{T} p_{Z} \leq 0,

(12)

which allows the objective function to be reduced by any step of the form

(x, s) + α p

, where

α > 0

. The vector

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

. If no finite step of the form

(x, s) + α p

(where

α > 0

) reaches a constraint not in the working set, the QP problem is unbounded and the function terminates immediately with

ifail = 2

(see Error Indicators and Warnings). Otherwise,

α

is defined as the maximum feasible step along

p

and a constraint active at

(x, s) + α 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

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

{(\begin{matrix} B & S \end{matrix})}^{T} = L U,

(13)

returning just the permutation

P

that makes

P L P^{T}

unit lower triangular. The pivot tolerance is set to require

{|P L P^{T}|}_{i j} \leq 2

, and the permutation is used to define

P

in (6). It can be shown that

‖Z‖

is likely to be little more than unity. Hence,

Z

should be well-conditioned regardless of the condition of

W

. This feature is applied at the beginning of the optimality phase if a potential

B - S

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)

by a small amount.

Suppose that the value of the optional parameter Feasibility Tolerance is

δ

. Over a period of

K

iterations (where

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

0.5 δ

δ

(in steps of

0.5 δ / 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

0.5 δ

If a problem requires more than

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

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

(where

α \leq α_{m}

) along

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

well-conditioned.

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 arguments 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 Description of the 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 $a$ , $i$ and $r$ denote options that take character, integer and real values respectively;
the default value is used whenever the condition $|i| \geq 100000000$ is satisfied and where the symbol $ε$ 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

Default

= 60

Every

i

th iteration after the most recent basis factorization, a numerical test is made to see if the current solution

(x, s)

satisfies the linear constraints

A x - s = 0

. If the largest element of the residual vector

r = A x - s

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

, the default value is used. If

i = 0

, the value

i = 99999999

is used and effectively no checks are made.

Crash Option

i

Default

= 2

Note that this option does not apply when

start ='W'

(see Arguments).

start ='C'

, an internal Crash procedure is used to select an initial basis from various rows and columns of the constraint matrix

(\begin{matrix} A & - I \end{matrix})

. The value of

i

determines which rows and columns are initially eligible for the basis, and how many times the Crash procedure is called. If

i = 0

, the all-slack basis

B = - I

is chosen. If

i = 1

, the Crash procedure is called once (looking for a triangular basis in all rows and columns of the linear constraint matrix

A

). If

i = 2

, the Crash procedure is called twice (looking at any equality constraints first followed by any inequality constraints). If

i < 0

i > 2

, the default value is used.

i = 1 ​ or ​ 2

, certain slacks on inequality rows are selected for the basis first. (If

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

, 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

Default

= 0.1

This value allows the Crash procedure to ignore certain ‘small’ nonzero elements in the constraint matrix

A

while searching for a triangular basis. For each column of

A

, if

a_{\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} \times r

When

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 \geq 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

Default

= 10000

This option is part of an anti-cycling procedure (see 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

. Over a period of

i

iterations, the feasibility tolerance actually used by nag_opt_qpconvex1_sparse_solve (e04nk) (i.e., the working feasibility tolerance) increases from

0.5 δ

δ

(in steps of

0.5 δ / 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

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).

i < 0

, the default value is used. If

i = 0

, the value

i = 99999999

is used and effectively no anti-cycling procedure is invoked.

Factorization Frequency

i

Default

= 100

i > 0

, at most

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

A x - s = 0

are satisfied. If necessary, the basis will be refactorized before the limit of

i

updates is reached. If

i \leq 0

, the default value is used.

Feasibility Tolerance

r

Default

= \max (10^{- 6}, \sqrt{ε})

r \geq ε

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

10^{- 5}

. If

r < ε

, 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

by a factor of

10

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

Default

= 10^{20}

r > 0

r

defines the ‘infinite’ bound

bigbnd

in the definition of the problem constraints. Any upper bound greater than or equal to

bigbnd

will be regarded as

+ \infty

(and similarly any lower bound less than or equal to

- bigbnd

will be regarded as

- \infty

). If

r \leq 0

, the default value is used.

Infinite Step Size

r

Default

= \max (bigbnd, 10^{20})

r > 0

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

during an iteration would exceed the value of

r

, the objective function is considered to be unbounded below in the feasible region. If

r \leq 0

, the default value is used.

Iteration Limit

i

Default

= \max (50, 5 (n + m))

Iters

Itns

The value of

i

specifies the maximum number of iterations allowed before termination. Setting

i = 0

and

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

, the default value is used.

List

Nolist

Default for nag_opt_qpconvex1_sparse_solve (e04nk)

= 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}

Default

= 100.0

LU Update Tolerance

r_{2}

Default

= 10.0

The values of

r_{1}

and

r_{2}

affect the stability and sparsity of the basis factorization

B = L U

, during refactorization and updates respectively. The lower triangular matrix

L

is a product of matrices of the form

(\begin{matrix} 1 \\ μ & 1 \end{matrix})

where the multipliers

μ

will satisfy

|μ| \leq r_{i}

. The default values of

r_{1}

and

r_{2}

usually strike a good compromise between stability and sparsity. For large and relatively dense problems, setting

r_{1}

and

r_{2}

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}

in the range

1 \leq r_{1} < 2

in order to achieve stability. If

r_{1} < 1

r_{2} < 1

, the default value is used.

LU Singularity Tolerance

r

Default

= ε^{0.67}

r > 0

r

defines the singularity tolerance used to guard against ill-conditioned basis matrices. Whenever the basis is refactorized, the diagonal elements of

U

are tested as follows. If

|u_{j j}| \leq r

|u_{j j}| < r \times \max_{i} |u_{i j}|

, the

j

th column of the basis is replaced by the corresponding slack variable. If

r \leq 0

, the default value is used.

Minimize

Default

Maximize

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)

and the other maximizes

- f (x)

, their solutions will be the same but the signs of the dual variables

π_{i}

and the reduced gradients

d_{j}

(see Main Iteration) will be reversed.

Monitoring File

i

Default

= - 1

i \geq 0

and

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

. If

i < 0

and/or

Print Level = 0

, the default value is used and hence no monitoring information is produced.

Optimality Tolerance

r

Default

= \max (10^{- 6}, \sqrt{ε})

r \geq ε

r

is used to judge the size of the reduced gradients

d_{j} = g_{j} - π^{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 \times \max (1, ‖π‖) \leq d_{j} \leq r \times \max (1, ‖π‖)

, and if

|d_{j}| \leq r \times \max (1, ‖π‖)

for any superbasic variables. If

r < ε

, the default value is used.

Partial Price

i

Default

= 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

). It reduces the work required for each pricing operation (i.e., when a nonbasic variable is selected to enter the basis). If

i = 1

, all columns of the constraint matrix

(\begin{matrix} A & - I \end{matrix})

are searched. If

i > 1

A

and

- I

are partitioned to give

i

roughly equal segments

A_{j}, K_{j}

, for

j = 1, 2, \dots, p

(modulo

p

). If the previous pricing search was successful on

A_{j - 1}, K_{j - 1}

, the next search begins on the segments

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}

, and so on. If

i \leq 0

, the default value is used.

Pivot Tolerance

r

Default

= ε^{0.67}

r > 0

r

is used to prevent columns entering the basis if they would cause the basis to become almost singular. If

r \leq 0

, the default value is used.

Print Level

i

The value of

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 Printed output (summary output at each iteration and the final solution) and Description of Monitoring Information (monitoring information at each iteration). Note that the summary output will not exceed

80

characters per line and that the monitoring information will not exceed

120

characters per line. If

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)):

$i$	Output
$0$	No output.
$1$	The final solution only.
$5$	One line of summary output for each iteration (no printout of the final solution).
$\geq 10$	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:

$i$	Output
$0$	No output.
$1$	The final solution only.
$5$	One long line of output for each iteration (no printout of the final solution).
$\geq 10$	The final solution and one long line of output for each iteration.
$\geq 20$	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 $Scale Option = 1$ or $2$ (see the description of the optional parameter Scale Option), basis factorization statistics and details of the initial basis resulting from the Crash procedure (if $start ='C'$ ; see Arguments).

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

Default

= 100 ε

Scale Option

i

Default

= 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}

and column scales

c_{j}

such that the scaled matrix coefficients

{\bar{a}}_{i j} = a_{i j} \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

{\bar{l}}_{j} = l_{j} / c_{j}

and

{\bar{u}}_{j} = u_{j} / c_{j}

respectively, where

c_{j} \equiv r_{j - n}

j > n

i = 0

, no scaling is performed. If

i = 1

, all rows and columns of the constraint matrix

A

are scaled. If

i = 2

, an additional scaling is performed that may be helpful when the solution

x

is large; it takes into account columns of

(\begin{matrix} A & - I \end{matrix})

that are fixed or have positive lower bounds or negative upper bounds. If

i < 0

i > 2

, the default value is used.

Scale Tolerance

r

Default

= 0.9

Note that this option does not apply when

Scale Option = 0

0 < r < 1

r

is used to control the number of scaling passes to be made through the constraint matrix

A

. At least

3

(and at most

10

) passes will be made. More precisely, let

a_{p}

denote the largest column ratio (i.e.,

\frac{'biggest' ​ element}{'smallest' ​ element}

in some sense) after the

p

th scaling pass through

A

. The scaling procedure is terminated if

a_{p} \geq a_{p - 1} \times r

for some

p \geq 3

. Thus, increasing the value of

r

from

0.9

0.99

(say) will probably increase the number of passes through

A

r \leq 0

r \geq 1

, the default value is used.

Superbasics Limit

i

Default

= \min (n_{H} + 1, n)

Note that this option does not apply to FP or LP problems.

The value of

i

specifies ‘how nonlinear’ you expect the QP problem to be. If

i \leq 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

\geq 10

and

Monitoring File \geq 0

, the following line of intermediate printout (

< 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 $A$ and $- I$ . 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 $α$ taken along the current search direction $p$ . The variables $x$ have just been changed to $x + α p$ . 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 $r$ th element of a vector $y$ satisfying $B y = a_{q}$ whenever $a_{q}$ (the $q$ th column of the constraint matrix $(\begin{matrix} A & - I \end{matrix})$ ) replaces the $r$ th column of the basis matrix $B$ . 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 $y$ 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 $x$ is not feasible, Sinf gives the sum of the magnitudes of constraint violations. If $x$ 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 $L$ . Immediately after a basis factorization $B = L U$ , this entry contains lenL. Further nonzeros are added to L when various columns of $B$ are later replaced. (Thus, L increases monotonically.)
U	is the number of nonzeros in the basis factor $U$ . Immediately after a basis factorization $B = L U$ , this entry contains lenU. As columns of $B$ are replaced, the matrix $U$ 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 $U$ . 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 $L + U$ and rerun nag_opt_qpconvex1_sparse_solve (e04nk) (possibly using $start ='W'$ ; see Arguments).
Norm rg	is $‖d_{S}‖$ , the Euclidean norm of the reduced gradient (see 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 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 \geq 20

and

Monitoring File \geq 0

, the following lines of intermediate printout (

< 120

characters) are produced on the unit number specified by optional parameter Monitoring File whenever the matrix

B

B_{S} = {(\begin{matrix} B & S \end{matrix})}^{T}

is factorized. Gaussian elimination is used to compute an

L U

factorization of

B

B_{S}

, where

P L P^{T}

is a lower triangular matrix and

P U Q

is an upper triangular matrix for some permutation matrices

P

and

Q

. The factorization is stabilized in the manner described under the optional parameter LU Factor Tolerance (

default value = 100.0

Factorize

is the factorization count.

Demand

is a code giving the reason for the present factorization as follows:

Code	Meaning
$0$	First $L U$ factorization.
$1$	The number of updates reached the value of the optional parameter Factorization Frequency.
$2$	The number of nonzeros in the updated factors has increased significantly.
$7$	Not enough storage to update factors.
$10$	Row residuals too large (see the description for the optional parameter Check Frequency).
$11$	Ill-conditioning has caused inconsistent results.

Iteration

is the iteration count.

Nonlinear

is the number of nonlinear variables in the current basis

B

(not printed if

B_{S}

is factorized).

Linear

is the number of linear variables in

B

(not printed if

B_{S}

is factorized).

Slacks

is the number of slack variables in

B

(not printed if

B_{S}

is factorized).

Elems

is the number of nonzeros in

B

(not printed if

B_{S}

is factorized).

Density

is the percentage nonzero density of

B

(not printed if

B_{S}

is factorized). More precisely,

Density = 100 \times Elems / {(Nonlinear + Linear + 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

4

, increase leniz and lenz and rerun nag_opt_qpconvex1_sparse_solve (e04nk) (possibly using

start ='W'

; see Arguments).

Merit

is the average Markowitz merit count for the elements chosen to be the diagonals of

P U Q

. Each merit count is defined to be

(c - 1) (r - 1)

, where

c

and

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

lenU

is the number of nonzeros in

U

Increase

is the percentage increase in the number of nonzeros in

L

and

U

relative to the number of nonzeros in

B

. More precisely,

Increase = 100 \times (lenL + lenU - Elems) / Elems

is the number of rows in the problem. Note that

m = Ut + Lt + bp

is the number of triangular rows of

B

at the top of

U

is the number of columns remaining when the density of the basis matrix being factorized reached

0.3

Lmax

is the maximum subdiagonal element in the columns of

L

. This will not exceed the value of the optional parameter LU Factor Tolerance.

Bmax

is the maximum nonzero element in

B

(not printed if

B_{S}

is factorized).

BSmax

is the maximum nonzero element in

B_{S}

(not printed if

B

is factorized).

Umax

is the maximum nonzero element in

U

, excluding elements of

B

that remain in

U

unchanged. (For example, if a slack variable is in the basis, the corresponding row of

B

will become a row of

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

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}

is factorized.

Umin

is the magnitude of the smallest diagonal element of

P U Q

(not printed if

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

), the ratio

\max (Bmax, Umax) / Umin

is an estimate of the condition number of

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 near-singularity by using slacks to replace columns of

B

that would have made Umin extremely small and the modified basis is refactorized.)

Growth is not printed if

B_{S}

is factorized.

is the number of triangular columns of

B

at the left of

L

is the size of the ‘bump’ or block to be factorized nontrivially after the triangular rows and columns of

B

have been removed.

is the number of columns remaining when the density of the basis matrix being factorized has reached

0.6

When

Print Level \geq 20

and

Monitoring File \geq 0

, the following lines of intermediate printout (

< 80

characters) are produced on the unit number specified by optional parameter Monitoring File whenever

start ='C'

(see Arguments). They refer to the number of columns selected by the Crash procedure during each of several passes through

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., $istate (j) = 3$ for some $j \leq n$ ). It will be a subset of the columns for which $istate (j) = 3$ 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 \geq 20

and

Monitoring File \geq 0

, the following lines of intermediate printout (

< 80

characters) are produced on the unit number specified by optional parameter Monitoring File. They refer to the elements of the names array (see Arguments).

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 right-hand 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

\geq 10

and

Monitoring File \geq 0

, the following lines of final printout (

< 120

characters) are produced on the unit number specified by optional parameter Monitoring File.

Let

a_{j}

denote the

j

th column of

A

, for

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

. (This is used internally to refer to

x_{j}

in the intermediate output.)

Column

gives the name of

x_{j}

State

A key is sometimes printed before State. Note that unless the optional parameter

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}

at the final iterate.

Obj Gradient

is the value of

g_{j}

at the final iterate. For FP problems,

g_{j}

is set to zero.

Lower Bound

is the lower bound specified for the variable. None indicates that

bl (j) \leq - bigbnd

Upper Bound

is the upper bound specified for the variable. None indicates that

bu (j) \geq bigbnd

Reduced Gradnt

is the value of

d_{j}

at the final iterate (see Main Iteration). For FP problems,

d_{j}

is set to zero.

m + j

is the value of

m + j

Let

v_{i}

denote the

i

th row of

A

, for

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

. (This is used internally to refer to

s_{i}

in the intermediate output.)

Row

gives the name of

ν_{i}

State

gives the state of

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}

is basic and SBS if inactive when

s_{i}

is superbasic).

A key is sometimes printed before State. Note that unless the optional parameter

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}

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) \leq - bigbnd

Upper Bound

is the upper bound specified for the variable. None indicates that

bu (j) \geq bigbnd

Dual Activity

is the value of the dual variable

π_{i}

(the Lagrange multiplier for

ν_{i}

; see Main Iteration). For FP problems,

π_{i}

is set to zero.

gives the index

i

of the

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, 64-bit version, 64-bit version)

Chapter Contents

Chapter Introduction

NAG Toolbox