Integer, Intent (In)	::	ltail, lp, la, lb
Integer, Intent (Inout)	::	ifail
Integer, Intent (Out)	::	ivalid(*)
Real (Kind=nag_wp), Intent (In)	::	p(lp), a(la), b(lb), tol
Real (Kind=nag_wp), Intent (Out)	::	g(*)
Character (1), Intent (In)	::	tail(ltail)

C Header Interface

#include <nag.h>

void	g01tff_ (const Integer ltail, const char tail[], const Integer lp, const double p[], const Integer la, const double a[], const Integer lb, const double b[], const double tol, double g[], Integer ivalid[], Integer ifail, const Charlen length_tail)

The routine may be called by the names g01tff or nagf_stat_inv_cdf_gamma_vector.

3 Description

The deviate,

g_{p_{i}}

, associated with the lower tail probability,

p_{i}

, of the gamma distribution with shape parameter

α_{i}

and scale parameter

β_{i}

, is defined as the solution to

P (G_{i} \leq g_{p_{i}} : α_{i}, β_{i}) = p_{i} = \frac{1}{{β_{i}}^{α_{i}} Γ (α_{i})} \int_{0}^{g_{p_{i}}} {e_{i}}^{- G_{i} / β_{i}} {G_{i}}^{α_{i} - 1} d G_{i}, 0 \leq g_{p_{i}} < \infty; ​ α_{i}, β_{i} > 0 .

The method used is described by Best and Roberts (1975) making use of the relationship between the gamma distribution and the

χ^{2}

-distribution.

Let

y_{i} = 2 \frac{g_{p_{i}}}{β_{i}}

. The required

y_{i}

is found from the Taylor series expansion

y_{i} = y_{0} + \sum_{r} \frac{C_{r} (y_{0})}{r!} {(\frac{E_{i}}{ϕ (y_{0})})}^{r},

where

y_{0}

is a starting approximation

$C_{1} (u_{i}) = 1$ ,
$C_{r + 1} (u_{i}) = (r Ψ + \frac{d}{d u_{i}}) C_{r} (u_{i})$ ,
$Ψ_{i} = \frac{1}{2} - \frac{α_{i} - 1}{u_{i}}$ ,
$E_{i} = p_{i} - \int_{0}^{y_{0}} ϕ_{i} (u_{i}) d u_{i}$ ,
$ϕ_{i} (u_{i}) = \frac{1}{2^{α_{i}} Γ (α_{i})} {e_{i}}^{- u_{i} / 2} {u_{i}}^{α_{i} - 1}$ .

For most values of

p_{i}

and

α_{i}

the starting value

y_{01} = 2 α_{i} {(z_{i} \sqrt{\frac{1}{9 α_{i}}} + 1 - \frac{1}{9 α_{i}})}^{3}

is used, where

z_{i}

is the deviate associated with a lower tail probability of

p_{i}

for the standard Normal distribution.

For

p_{i}

close to zero,

y_{02} = {(p_{i} α_{i} 2^{α_{i}} Γ (α_{i}))}^{1 / α_{i}}

is used.

For large

p_{i}

values, when

y_{01} > 4.4 α_{i} + 6.0

y_{03} = - 2 [\ln (1 - p_{i}) - (α_{i} - 1) \ln (\frac{1}{2} y_{01}) + \ln (Γ (α_{i}))]

is found to be a better starting value than

y_{01}

For small

α_{i}

(α_{i} \leq 0.16)

p_{i}

is expressed in terms of an approximation to the exponential integral and

y_{04}

is found by Newton–Raphson iterations.

Seven terms of the Taylor series are used to refine the starting approximation, repeating the process if necessary until the required accuracy is obtained.

The input arrays to this routine are designed to allow maximum flexibility in the supply of vector arguments by re-using elements of any arrays that are shorter than the total number of evaluations required. See Section 2.6 in the G01 Chapter Introduction for further information.

4 References

Best D J and Roberts D E (1975) Algorithm AS 91. The percentage points of the

χ^{2}

distribution Appl. Statist. 24 385–388

5 Arguments

1: $ltail$ – Integer Input

On entry: the length of the array tail.

Constraint:

ltail > 0

2: $tail (ltail)$ – Character(1) array Input

On entry: indicates which tail the supplied probabilities represent. For

j = ((i - 1) mod ltail) + 1

, for

i = 1, 2, \dots, \max (ltail, lp, la, lb)

$tail (j) ='L'$: The lower tail probability, i.e., $p_{i} = P (G_{i} \leq g_{p_{i}} : α_{i}, β_{i})$ .
$tail (j) ='U'$: The upper tail probability, i.e., $p_{i} = P (G_{i} \geq g_{p_{i}} : α_{i}, β_{i})$ .

Constraint:

tail (j) ='L'

'U'

, for

j = 1, 2, \dots, ltail

3: $lp$ – Integer Input

On entry: the length of the array p.

Constraint:

lp > 0

4: $p (lp)$ – Real (Kind=nag_wp) array Input

On entry:

p_{i}

, the probability of the required gamma distribution as defined by tail with

p_{i} = p (j)

j = ((i - 1) mod lp) + 1

Constraints:

if $tail (k) ='L'$ , $0.0 \leq p (j) < 1.0$ ;
otherwise $0.0 < p (j) \leq 1.0$ .

Where

k = (i - 1) mod ltail + 1

and

j = (i - 1) mod lp + 1

5: $la$ – Integer Input

On entry: the length of the array a.

Constraint:

la > 0

6: $a (la)$ – Real (Kind=nag_wp) array Input

On entry:

α_{i}

, the first parameter of the required gamma distribution with

α_{i} = a (j)

j = ((i - 1) mod la) + 1

Constraint:

0.0 < a (j) \leq 10^{6}

, for

j = 1, 2, \dots, la

7: $lb$ – Integer Input

On entry: the length of the array b.

Constraint:

lb > 0

8: $b (lb)$ – Real (Kind=nag_wp) array Input

On entry:

β_{i}

, the second parameter of the required gamma distribution with

β_{i} = b (j)

j = ((i - 1) mod lb) + 1

Constraint:

b (j) > 0.0

, for

j = 1, 2, \dots, lb

9: $tol$ – Real (Kind=nag_wp) Input

On entry: the relative accuracy required by you in the results. If g01tff is entered with tol greater than or equal to

1.0

or less than

10 \times machine precision

(see x02ajf), the value of

10 \times machine precision

is used instead.

10: $g (*)$ – Real (Kind=nag_wp) array Output

Note: the dimension of the array g must be at least

\max (ltail, lp, la, lb)

On exit:

g_{p_{i}}

, the deviates for the gamma distribution.

11: $ivalid (*)$ – Integer array Output

Note: the dimension of the array ivalid must be at least

\max (ltail, lp, la, lb)

On exit:

ivalid (i)

indicates any errors with the input arguments, with

$ivalid (i) = 0$: No error.
$ivalid (i) = 1$: On entry, invalid value supplied in tail when calculating $g_{p_{i}}$ .
$ivalid (i) = 2$: On entry, invalid value for $p_{i}$ .
$ivalid (i) = 3$: On entry, $α_{i} \leq 0.0$ , or, $α_{i} > 10^{6}$ , or, $β_{i} \leq 0.0$ .
$ivalid (i) = 4$: $p_{i}$ is too close to $0.0$ or $1.0$ to enable the result to be calculated.
$ivalid (i) = 5$: The solution has failed to converge. The result may be a reasonable approximation.

12: $ifail$ – Integer Input/Output

On entry: ifail must be set to

0

- 1

1

to set behaviour on detection of an error; these values have no effect when no error is detected.

A value of

0

causes the printing of an error message and program execution will be halted; otherwise program execution continues. A value of

- 1

means that an error message is printed while a value of

1

means that it is not.

If halting is not appropriate, the value

- 1

1

is recommended. If message printing is undesirable, then the value

1

is recommended. Otherwise, the value

- 1

is recommended since useful values can be provided in some output arguments even when

ifail \neq 0

on exit. When the value $- 1$ or $1$ is used it is essential to test the value of ifail on exit.

On exit:

ifail = 0

unless the routine detects an error or a warning has been flagged (see Section 6).

6 Error Indicators and Warnings

If on entry

ifail = 0

- 1

, explanatory error messages are output on the current error message unit (as defined by x04aaf).

Errors or warnings detected by the routine:

Note: in some cases g01tff may return useful information.

$ifail = 1$: On entry, at least one value of tail, p, a, or b was invalid.
Check ivalid for more information.

$ifail = 2$: On entry, $array size = 〈value〉$ .
Constraint: $ltail > 0$ .

$ifail = 3$: On entry, $array size = 〈value〉$ .
Constraint: $lp > 0$ .

$ifail = 4$: On entry, $array size = 〈value〉$ .
Constraint: $la > 0$ .

$ifail = 5$: On entry, $array size = 〈value〉$ .
Constraint: $lb > 0$ .

$ifail = - 99$: An unexpected error has been triggered by this routine. Please contact NAG.
See Section 7 in the Introduction to the NAG Library FL Interface for further information.

$ifail = - 399$: Your licence key may have expired or may not have been installed correctly.
See Section 8 in the Introduction to the NAG Library FL Interface for further information.

$ifail = - 999$: Dynamic memory allocation failed.
See Section 9 in the Introduction to the NAG Library FL Interface for further information.

7 Accuracy

In most cases the relative accuracy of the results should be as specified by tol. However, for very small values of

α_{i}

or very small values of

p_{i}

there may be some loss of accuracy.

8 Parallelism and Performance

g01tff is not threaded in any implementation.

9 Further Comments

None.

10 Example

This example reads lower tail probabilities for several gamma distributions, and calculates and prints the corresponding deviates until the end of data is reached.

g01tf: FL CL CPP AD

NAG FL Interfaceg01tff (inv_​cdf_​gamma_​vector)

▸▿ Contents

1 Purpose

2 Specification

3 Description

4 References

5 Arguments

6 Error Indicators and Warnings

7 Accuracy

8 Parallelism and Performance

9 Further Comments

10 Example

10.1 Program Text

10.2 Program Data

10.3 Program Results

NAG FL Interface
g01tff (inv_cdf_gamma_vector)