# NAG Library Routine Document

## 1Purpose

g13caf calculates the smoothed sample spectrum of a univariate time series using one of four lag windows – rectangular, Bartlett, Tukey or Parzen window.

## 2Specification

Fortran Interface
 Subroutine g13caf ( nx, mtx, px, iw, mw, ic, nc, c, kc, l, lg, nxg, xg, ng,
 Integer, Intent (In) :: nx, mtx, iw, mw, ic, nc, kc, l, lg, nxg Integer, Intent (Inout) :: ifail Integer, Intent (Out) :: ng Real (Kind=nag_wp), Intent (In) :: px Real (Kind=nag_wp), Intent (Inout) :: c(nc), xg(nxg) Real (Kind=nag_wp), Intent (Out) :: stats(4)
#include nagmk26.h
 void g13caf_ (const Integer *nx, const Integer *mtx, const double *px, const Integer *iw, const Integer *mw, const Integer *ic, const Integer *nc, double c[], const Integer *kc, const Integer *l, const Integer *lg, const Integer *nxg, double xg[], Integer *ng, double stats[], Integer *ifail)

## 3Description

The smoothed sample spectrum is defined as
 $f^ω=12π C0+2∑k=1 M-1wkCkcosωk ,$
where $M$ is the window width, and is calculated for frequency values
 $ωi=2πiL, i=0,1,…,L/2,$
where $\left[\right]$ denotes the integer part.
The autocovariances ${C}_{k}$ may be supplied by you, or constructed from a time series ${x}_{1},{x}_{2},\dots ,{x}_{n}$, as
 $Ck=1n∑t=1 n-kxtxt+k,$
the fast Fourier transform (FFT) being used to carry out the convolution in this formula.
The time series may be mean or trend corrected (by classical least squares), and tapered before calculation of the covariances, the tapering factors being those of the split cosine bell:
 $121-cosπ t-12/T, 1≤t≤T 121-cosπ n-t+12/T, n+ 1-T≤t≤n 1, otherwise,$
where $T=\left[\frac{np}{2}\right]$ and $p$ is the tapering proportion.
The smoothing window is defined by
 $wk=W kM , k≤M-1,$
which for the various windows is defined over $0\le \alpha <1$ by
rectangular:
 $Wα=1$
Bartlett:
 $Wα = 1-α$
Tukey:
 $Wα=121+cosπα$
Parzen:
 $Wα= 1- 6α2+ 6α3, 0≤α≤12 Wα= 2 1-α 3, 12<α< 1.$
The sampling distribution of $\stackrel{^}{f}\left(\omega \right)$ is approximately that of a scaled ${\chi }_{d}^{2}$ variate, whose degrees of freedom $d$ is provided by the routine, together with multiplying limits $mu$, $ml$ from which approximate $95%$ confidence intervals for the true spectrum $f\left(\omega \right)$ may be constructed as $\left[ml×\stackrel{^}{f}\left(\omega \right),mu×\stackrel{^}{f}\left(\omega \right)\right]$. Alternatively, log $\stackrel{^}{f}\left(\omega \right)$ may be returned, with additive limits.
The bandwidth $b$ of the corresponding smoothing window in the frequency domain is also provided. Spectrum estimates separated by (angular) frequencies much greater than $b$ may be assumed to be independent.
Bloomfield P (1976) Fourier Analysis of Time Series: An Introduction Wiley
Jenkins G M and Watts D G (1968) Spectral Analysis and its Applications Holden–Day

## 5Arguments

1:     $\mathbf{nx}$ – IntegerInput
On entry: $n$, the length of the time series.
Constraint: ${\mathbf{nx}}\ge 1$.
2:     $\mathbf{mtx}$ – IntegerInput
On entry: if covariances are to be calculated by the routine (${\mathbf{ic}}=0$), mtx must specify whether the data are to be initially mean or trend corrected.
${\mathbf{mtx}}=0$
For no correction.
${\mathbf{mtx}}=1$
For mean correction.
${\mathbf{mtx}}=2$
For trend correction.
Constraint: if ${\mathbf{ic}}=0$, $0\le {\mathbf{mtx}}\le 2$
If covariances are supplied (${\mathbf{ic}}\ne 0$), mtx is not used.
3:     $\mathbf{px}$ – Real (Kind=nag_wp)Input
On entry: if covariances are to be calculated by the routine (${\mathbf{ic}}=0$), px must specify the proportion of the data (totalled over both ends) to be initially tapered by the split cosine bell taper.
If covariances are supplied $\left({\mathbf{ic}}\ne 0\right)$, px must specify the proportion of data tapered before the supplied covariances were calculated and after any mean or trend correction. px is required for the calculation of output statistics. A value of $0.0$ implies no tapering.
Constraint: $0.0\le {\mathbf{px}}\le 1.0$.
4:     $\mathbf{iw}$ – IntegerInput
On entry: the choice of lag window.
${\mathbf{iw}}=1$
Rectangular.
${\mathbf{iw}}=2$
Bartlett.
${\mathbf{iw}}=3$
Tukey.
${\mathbf{iw}}=4$
Parzen.
Constraint: $1\le {\mathbf{iw}}\le 4$.
5:     $\mathbf{mw}$ – IntegerInput
On entry: $M$, the ‘cut-off’ point of the lag window. Windowed covariances at lag $M$ or greater are zero.
Constraint: $1\le {\mathbf{mw}}\le {\mathbf{nx}}$.
6:     $\mathbf{ic}$ – IntegerInput
On entry: indicates whether covariances are to be calculated in the routine or supplied in the call to the routine.
${\mathbf{ic}}=0$
Covariances are to be calculated.
${\mathbf{ic}}\ne 0$
Covariances are to be supplied.
7:     $\mathbf{nc}$ – IntegerInput
On entry: the number of covariances to be calculated in the routine or supplied in the call to the routine.
Constraint: ${\mathbf{mw}}\le {\mathbf{nc}}\le {\mathbf{nx}}$.
8:     $\mathbf{c}\left({\mathbf{nc}}\right)$ – Real (Kind=nag_wp) arrayInput/Output
On entry: if ${\mathbf{ic}}\ne 0$, c must contain the nc covariances for lags from $0$ to $\left({\mathbf{nc}}-1\right)$, otherwise c need not be set.
On exit: if ${\mathbf{ic}}=0$, c will contain the nc calculated covariances.
If ${\mathbf{ic}}\ne 0$, the contents of c will be unchanged.
9:     $\mathbf{kc}$ – IntegerInput
On entry: if ${\mathbf{ic}}=0$, kc must specify the order of the fast Fourier transform (FFT) used to calculate the covariances.
If ${\mathbf{ic}}\ne 0$, that is covariances are supplied, kc is not used.
Constraint: ${\mathbf{kc}}\ge {\mathbf{nx}}+{\mathbf{nc}}$.
10:   $\mathbf{l}$ – IntegerInput
On entry: $L$, the frequency division of the spectral estimates as $\frac{2\pi }{L}$. Therefore it is also the order of the FFT used to construct the sample spectrum from the covariances.
Constraint: ${\mathbf{l}}\ge 2×{\mathbf{mw}}-1$.
11:   $\mathbf{lg}$ – IntegerInput
On entry: indicates whether unlogged or logged spectral estimates and confidence limits are required.
${\mathbf{lg}}=0$
Unlogged.
${\mathbf{lg}}\ne 0$
Logged.
12:   $\mathbf{nxg}$ – IntegerInput
On entry: the dimension of the array xg as declared in the (sub)program from which g13caf is called.
Constraints:
• if ${\mathbf{ic}}=0$, ${\mathbf{nxg}}\ge \mathrm{max}\phantom{\rule{0.125em}{0ex}}\left({\mathbf{kc}},{\mathbf{l}}\right)$;
• if ${\mathbf{ic}}\ne 0$, ${\mathbf{nxg}}\ge {\mathbf{l}}$.
13:   $\mathbf{xg}\left({\mathbf{nxg}}\right)$ – Real (Kind=nag_wp) arrayInput/Output
On entry: if the covariances are to be calculated, then xg must contain the nx data points. If covariances are supplied, xg may contain any values.
On exit: contains the ng spectral estimates, $\stackrel{^}{f}\left({\omega }_{\mathit{i}}\right)$, for $\mathit{i}=0,1,\dots ,\left[L/2\right]$ in ${\mathbf{xg}}\left(1\right)$ to ${\mathbf{xg}}\left({\mathbf{ng}}\right)$ respectively (logged if ${\mathbf{lg}}=1$). The elements ${\mathbf{xg}}\left(\mathit{i}\right)$, for $\mathit{i}={\mathbf{ng}}+1,\dots ,{\mathbf{nxg}}$ contain $0.0$.
14:   $\mathbf{ng}$ – IntegerOutput
On exit: the number of spectral estimates, $\left[L/2\right]+1$, in xg.
15:   $\mathbf{stats}\left(4\right)$ – Real (Kind=nag_wp) arrayOutput
On exit: four associated statistics. These are the degrees of freedom in ${\mathbf{stats}}\left(1\right)$, the lower and upper $95%$ confidence limit factors in ${\mathbf{stats}}\left(2\right)$ and ${\mathbf{stats}}\left(3\right)$ respectively (logged if ${\mathbf{lg}}=1$), and the bandwidth in ${\mathbf{stats}}\left(4\right)$.
16:   $\mathbf{ifail}$ – IntegerInput/Output
On entry: ifail must be set to $0$, $-1\text{​ or ​}1$. If you are unfamiliar with this argument you should refer to Section 3.4 in How to Use the NAG Library and its Documentation for details.
For environments where it might be inappropriate to halt program execution when an error is detected, the value $-1\text{​ or ​}1$ is recommended. If the output of error messages is undesirable, then the value $1$ is recommended. Otherwise, if you are not familiar with this argument, the recommended value is $0$. When the value $-\mathbf{1}\text{​ or ​}\mathbf{1}$ is used it is essential to test the value of ifail on exit.
On exit: ${\mathbf{ifail}}={\mathbf{0}}$ unless the routine detects an error or a warning has been flagged (see Section 6).

## 6Error Indicators and Warnings

If on entry ${\mathbf{ifail}}=0$ or $-1$, explanatory error messages are output on the current error message unit (as defined by x04aaf).
Errors or warnings detected by the routine:
${\mathbf{ifail}}=1$
On entry, ${\mathbf{ic}}=0$ and ${\mathbf{mtx}}<0$: ${\mathbf{mtx}}=〈\mathit{\text{value}}〉$.
On entry, ${\mathbf{ic}}=0$ and ${\mathbf{mtx}}>2$: ${\mathbf{mtx}}=〈\mathit{\text{value}}〉$.
On entry, ${\mathbf{iw}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{iw}}=1$, $2$, $3$ or $4$.
On entry, ${\mathbf{mw}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{mw}}\ge 1$.
On entry, ${\mathbf{mw}}=〈\mathit{\text{value}}〉$ and ${\mathbf{nx}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{mw}}\le {\mathbf{nx}}$.
On entry, ${\mathbf{nc}}=〈\mathit{\text{value}}〉$ and ${\mathbf{mw}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{nc}}\ge {\mathbf{mw}}$.
On entry, ${\mathbf{nc}}=〈\mathit{\text{value}}〉$ and ${\mathbf{nx}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{nc}}\le {\mathbf{nx}}$.
On entry, ${\mathbf{nx}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{nx}}\ge 1$.
On entry, ${\mathbf{nxg}}=〈\mathit{\text{value}}〉$, ${\mathbf{kc}}=〈\mathit{\text{value}}〉$ and ${\mathbf{l}}=〈\mathit{\text{value}}〉$.
Constraint: if ${\mathbf{ic}}=0$, ${\mathbf{nxg}}\ge \mathrm{max}\phantom{\rule{0.125em}{0ex}}\left({\mathbf{kc}},{\mathbf{l}}\right)$.
On entry, ${\mathbf{nxg}}=〈\mathit{\text{value}}〉$ and ${\mathbf{l}}=〈\mathit{\text{value}}〉$.
Constraint: if ${\mathbf{ic}}\ne 0$, ${\mathbf{nxg}}\ge {\mathbf{l}}$.
On entry, ${\mathbf{px}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{px}}\ge 0.0$.
On entry, ${\mathbf{px}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{px}}\le 1.0$.
${\mathbf{ifail}}=2$
On entry, ${\mathbf{kc}}=〈\mathit{\text{value}}〉$, ${\mathbf{nx}}=〈\mathit{\text{value}}〉$ and ${\mathbf{nc}}=〈\mathit{\text{value}}〉$.
Constraint: if ${\mathbf{ic}}=0$, ${\mathbf{kc}}\ge \left({\mathbf{nx}}+{\mathbf{nc}}\right)$.
${\mathbf{ifail}}=3$
On entry, ${\mathbf{l}}=〈\mathit{\text{value}}〉$ and ${\mathbf{mw}}=〈\mathit{\text{value}}〉$.
Constraint: ${\mathbf{l}}\ge 2×{\mathbf{mw}}-1$.
${\mathbf{ifail}}=4$
One or more spectral estimates are negative.
Unlogged spectral estimates are returned in xg, and the degrees of freedom, unloged confidence limit factors and bandwidth in stats.
${\mathbf{ifail}}=5$
The calculation of confidence limit factors has failed.
Spectral estimates (logged if requested) are returned in xg, and degrees of freedom and bandwidth in stats.
${\mathbf{ifail}}=-99$
See Section 3.9 in How to Use the NAG Library and its Documentation for further information.
${\mathbf{ifail}}=-399$
Your licence key may have expired or may not have been installed correctly.
See Section 3.8 in How to Use the NAG Library and its Documentation for further information.
${\mathbf{ifail}}=-999$
Dynamic memory allocation failed.
See Section 3.7 in How to Use the NAG Library and its Documentation for further information.

## 7Accuracy

The FFT is a numerically stable process, and any errors introduced during the computation will normally be insignificant compared with uncertainty in the data.

## 8Parallelism and Performance

g13caf is threaded by NAG for parallel execution in multithreaded implementations of the NAG Library.
g13caf makes calls to BLAS and/or LAPACK routines, which may be threaded within the vendor library used by this implementation. Consult the documentation for the vendor library for further information.
Please consult the X06 Chapter Introduction for information on how to control and interrogate the OpenMP environment used within this routine. Please also consult the Users' Note for your implementation for any additional implementation-specific information.

g13caf carries out two FFTs of length kc to calculate the covariances and one FFT of length l to calculate the sample spectrum. The time taken by the routine for an FFT of length $n$ is approximately proportional to $n\mathrm{log}\left(n\right)$ (but see Section 9 in c06paf for further details).

## 10Example

This example reads a time series of length $256$. It selects the mean correction option, a tapering proportion of $0.1$, the Parzen smoothing window and a cut-off point for the window at lag $100$. It chooses to have $100$ auto-covariances calculated and unlogged spectral estimates at a frequency division of $2\pi /200$. It then calls g13caf to calculate the univariate spectrum and statistics and prints the autocovariances and the spectrum together with its $95%$ confidence multiplying limits.

### 10.1Program Text

Program Text (g13cafe.f90)

### 10.2Program Data

Program Data (g13cafe.d)

### 10.3Program Results

Program Results (g13cafe.r)

© The Numerical Algorithms Group Ltd, Oxford, UK. 2017