Overflow may occur if the routine attempts to scale the polynomial coefficients.

In rare circumstances overflow may be observed if $\mathbf{scal}=\mathrm{.TRUE.}$.

Non-critical

16

Set argument $\mathbf{scal}=\mathrm{.FALSE.}$.

d02ngf can fail on stationary problems.

d02ngf can fail on stationary problems..

Critical

23

27.3

d06acf returns $\mathbf{ifail}=\mathbf{2}$error for some boundary meshes due to an internal scaling issue.

d06acf returns $\mathbf{ifail}=\mathbf{2}$error for some boundary meshes due to an internal scaling issue.

Non-critical

20

28.4

Scale input boundary mesh prior to calling d06acf so that $\mathbf{}\mathbf{}=0$and $\mathbf{}\mathbf{}=1$.

e01shf will occasionally incorrectly identify a point as being outside the region defined by the interpolant.

e01shf will occasionally incorrectly identify a point as being outside the region defined by the interpolant. This leads to the function value being extrapolated rather than interpolated and can lead to incorrect results.

Non-critical

26.0

27.1

None.

No check that a mandatory call to the initialization routine has been made.

Whilst it is necessary to call initialization routine e04wbf prior to calling the named e04 routines, no software check is made to ensure that this has happened.

Non-critical

20

Ensure that initialization routine e04wbf has been called.

In some cases, the solver hangs indefinitely at an iteration.

In some cases, when certain variable lower bounds are not present, the solver might hang indefinitely due to an infinite loop during the line search stage.

Critical

27.1

30.3

None.

and $\mathbf{rinfo}$were not correctly filled by the presolver.

The arrays $\mathbf{stats}$and $\mathbf{rinfo}$were not correctly filled when the problem was entirely solved by the presolver. It now returns the correct values.

Non-critical

26.1

27

Don't rely on $\mathbf{rinfo}\left(1\right),\mathbf{rinfo}\left(2\right)$to hold the primal and dual objective in this case and recompute it as $c^{\prime }x$and $by$, respectively.

e04mtf does not report the correct solution when $3$or more columns are proportional to each other in the constraint matrix.

e04mtf does not report the correct solution when $3$or more columns are proportional to each other in the constraint matrix. In such a case, the solution reported may be infeasible.

Non-critical

26.1

27

A workaround is to disable the more complex presolve operations by setting the optional parameter $\mathbf{LP\; Presolve}=\mathrm{BASIC}$. This may slow down the solver depending on the problem.

In some very rare cases, the solution reported presents big violations on a small number of linear constraints.

In some very rare cases, the solution reported presents big violations on a small number of linear constraints.

Non-critical

26.1

27.1

A workaround is to deactivate the more complex presolver operations with the optional parameter $\mathbf{LP\; Presolve}=\mathrm{BASIC}$.

In some very rare cases, e04mtf reports problem infeasibility for a feasible problem.

In some very rare cases, the solver reports problem infeasibility when there are numerical difficulties.

Non-critical

26.1

28.6

Unfortunately there is no convenient workaround.

In some very rare cases, at presolve phase the solver declares unboundedness on a bounded problem.

In some very rare cases, especially when there are a large amount of singleton variables, the solver might report unbounded error message on a bounded problem.

Non-critical

26.1

29.0

A workaround is to deactivate the more complex presolver operations with the optional parameter $\mathbf{LP\; Presolve}=\mathrm{BASIC}$. This may slow down the solver depending on the problem.

In some cases the solver declares dual infeasibility during presolving for feasible and bounded problems.

For a bounded problem, the solver reports dual infeasibility during dominated columns removal at the presolving stage.

Non-critical

26.1

29.1

A workaround is to deactivate the more complex presolver operations with the optional parameter $\mathbf{LP\; Presolve}=\mathrm{BASIC}$. This may slow down the solver depending on the problem.

Infeasible bounds defined by e04rjf of a variable are ignored and infeasibility is not reported.

When infeasible bounds are defined by e04rjf for a variable, instead of reporting problem infeasibility, the bounds are overridden and wrong solution may be reported.

Non-critical

26.1

27.1

A workaround is to deactivate the more complex presolver operations with the optional parameter $\mathbf{LP\; Presolve}=\mathrm{BASIC}$for e04mtf and $\mathbf{SOCP\; Presolve}=\mathrm{BASIC}$for e04ptf.

Optional parameters $\mathbf{List}$and $\mathbf{Nolist}$are not handled correctly.

Routines e04nrf, e04vkf and e04wef do not handle optional parameters $\mathbf{List}$and $\mathbf{Nolist}$correctly. Specifying $\mathbf{List}$does not alter the behaviour of subsequent routines in the suite, and specifying $\mathbf{Nolist}$erroneously reports an error.

Non-critical

8

27.3

Routine e04nsf should be used instead to set optional parameters $\mathbf{List}$or $\mathbf{Nolist}$.

Information about the last constraint might not be printed.

If the problem has a nonlinear objective function without a linear part and $\mathbf{objrow}<\mathbf{nf}$, the last constraint is not printed in the final information about the solution (Rows section).

Non-critical

21

26

None.

When $\mathbf{ifail}=\mathbf{4}$is returned, the expected value of $\mathbf{nvar}$in the error message may be wrongly reported.

When $\mathbf{ifail}=\mathbf{4}$is returned, the expected value of $\mathbf{nvar}$in the error message may be wrongly reported. This incorrect behaviour may occur when the internal solver works with a different number of variables than what is declared during handle initialization, e.g., when some variables are fixed by the user.

Non-critical

28.3

30.2

None.

Multithreaded versions of the routines f11bef, f11bsf, f11gef and f11gsf may produce slightly different results when run on multiple threads.

Multithreaded versions of the routines f11bef, f11bsf, f11gef and f11gsf may produce slightly different results when run on multiple threads, e.g., the number of iterations to solution and the computed matrix norms and termination criteria reported by the associated monitoring routines. A bug affecting f11bef and f11gef has been fixed, and parallel vector dot products have been modified in all routines to improve consistency of results.

Non-critical

19

27.1

None.

f16rbf and f16ubf return $0$if $\mathbf{kl}$or $\mathbf{ku}$is $0$, instead of the correct norm. $\mathbf{ldab}$is incorrectly forced to be at least $\mathbf{m}$when $m=n$.

f16rbf and f16ubf mistakenly make a quick return if $\mathbf{kl}$or $\mathbf{ku}$is $0$, instead of computing the correct value for the requested norm. Also, $\mathbf{ldab}$is incorrectly forced to be at least $\mathbf{m}$when $m=n$.

Critical

23

27.3

None.

If the first row or column of the weight matrix $\mathbf{h}$consists only of zeros, then the routine will fail to find the nearest correlation matrix for that weight matrix.

An error can occur when the first row or column of the weight matrix $\mathbf{h}$consists only of zeros, then the routine will fail to find the nearest correlation matrix for that weight matrix.

Critical

25

30.1

The workaround is to apply a small weight (relative to other non-zero weights given in $\mathbf{h}$) to one of the elements in the first row/column.

A segmentation fault is likely to occur if a model with multiple random statements is supplied to the routine, where at least one of those statements does not have a $\mathbf{Subject}$term.

A segmentation fault is likely to occur if a model with multiple random statements is supplied to the routine, where at least one of those statements does not have a $\mathbf{Subject}$term.

V1 + V2 / SUBJECT = V3
V4 + V5 / SUBJECT = V6

V1 + V2
V4 + V5 / SUBJECT = V6

V1 + V2

Critical

27

27.1

A workaround to this issue is to always supply a $\mathbf{Subject}$term. If the required model is of the form:

V1 + V2
V4 + V5 / SUBJECT = V6

then you can specify an equivalent model by using:

V1 + V2 / SUBJECT = DUMMY
V4 + V5 / SUBJECT = V6

where the variable DUMMY has a single level, and always takes a value of one. This will require adding the variable DUMMY to your dataset.

The wrong value for $\mathbf{p}$is returned when $\mathbf{aa2}$is large.

In g08ckf and g08clf the value returned for the upper tail probability $\mathbf{p}$is wrong when the calculated Anderson-Darling test statistic $\mathbf{aa2}$is large. In the case of g08ckf, when $\mathbf{aa2}>153.4677$the returned value of $\mathbf{p}$should be zero; in the case of g08clf, when $\mathbf{aa2}>10.03$the returned value of $\mathbf{p}$should be $\le \exp \left(-14.360135\right)$.

Critical

23

Workaround for g08ckf:

Call g08ckf(...)
If (aa2 > 153.4677d0) p = 0.0d0

Workaround for g08clf:

Call g08clf(...)
If (aa2 > 10.03d0) p = exp(-14.360135d0)

g13faf may return a negative value as the estimate of the last $\beta$parameter (i.e., ${\beta }_p$) for a subset of models.

g13faf can result in a negative value for the estimate of the last $\beta$parameter (i.e., ${\beta }_p$) or, if $p=0$, the last $\alpha$parameter (i.e., ${\alpha }_q$).

Critical

20

27

None

When $\mathbf{what}=\text{'V'}$the information returned in $\mathbf{plab}$and/or $\mathbf{vinfo}$may be incorrect.

The information returned in $\mathbf{plab}$and/or $\mathbf{vinfo}$may be incorrect in cases where $\mathbf{what}=\text{'V'}$and the underlying linear mixed effects regression model has a random variable, with a single level (so either binary or continuous), that only takes the value zero.

Non-critical

27.0

The work around is to drop the term from the model, as it does not contribute. For example, if the random part of your model was specified as: V1 + V2 / SUBJECT=V3 and the variable V2 was a continuous variable, that only takes a value of zero in the data, then this is equivalent to re-specifying the model using: V1 / SUBJECT=V3.

$\mathbf{rinfo}$ was not correctly filled with the primal objective value and the dual objective bound when fixed variable presents.

Primal objective value and the dual objective bound were not returned correctly in $\mathbf{rinfo}$ due to the removal of fixed variables from the model. It now returns the correct values.

Non-critical

29.3

31.0

Recompute primal objective value as $c\prime x$.

In some cases, solver returns inaccurate solution when there are binary variables.

When there are binary variables, the accuracy of the solution will be influenced by the initial values. The solver could return inaccurate solution.

Non-critical

25

29.1

The work around is to use integer variables with $0$and $1$as lower and upper bound, respectively.

Thread Local Storage default limit was exceeded for delay loaded shared library.

A fair amount of thread local storage had been allocated by an auxiliary routine which has now been updated to use a very small amount of thread local storage. Prior to the update, this only affected the case where the shared version of the Nag Library was delay loaded, since this assumed a small default maximum amount of thread local storage, which was in fact exceeded.

Non-critical

26.1

28.6

None.

On Linux, static linking of the nAG Library using a recent Intel Fortran compiler results in a link error due to multiple defines.

(This error report applies to Linux library NLL6I only).

Critical

The recommended workaround is to link to the shared Library when this problem presents itself.