You are on page 1of 14

Nuclear Instruments and Methods 178 (1980) 499-512

© North-Holland Publishing Company

ANALYTICAL FUNCTIONS F O R FITTING PEAKS FROM Ge SEMICONDUCTOR DETECTORS *

R.G. HELMER and M.A. LEE


Idaho National Engineering Laboratory, EG & G Idaho, Inc., Idaho Falls, Idaho 83415, USA

Received 28 May 1980

A comparison of the quality-of-fit to ~/-ray peaks has been made for several analytical functions. The data are for 1332 keV
peaks measured on three Ge(Li) and Ge detectors with volumes of about 13, 65 and 120 cm 3. The analytical functions tested,
most previously proposed, have 0, 1 or 2 additive components to represent the low-energy tailing. Since non of the functions
represent the physical processes in the detection system, these analyses yield the expected result that, in general, the functions
with the most parameters give the best fits. The variation of the peak areas with some of the parameters of the tailing functions is
investigated.

1. Introduction tions (especially those o f SAMPO [5] and


GAMANAL [6]) are widely used. In several compari-
In determinations o f the emission rates of ")'-ray sons [ 7 - 9 ] o f the quality o f the fits to specific peaks,
lines made from 33-ray spectra measured with Ge semi- the advantage of the more complex (i.e., those with
conductor detectors, the precision is generally limited
10 5 i i ~ I
by the ability to define and compute consistent peak
A
areas. For single, well-resolved peaks, simple methods
for peak area determinations are generally adequate.
Summation of the counts in the region of the peak
can be used and uncertainties in the 33-ray emission
rates o f about 0.5% have been reported [ 1 - 3 ] . Also,
10 4
fits to the peak counts with a Gaussian function are
often used and uncertainties in the emission rates o f
about 1% are reported [4]. However, both the sum-
mation method and the simple Gaussian fitting func-
tion are only able to obtain precise areas for single
peaks. This is clear from the 33-ray peaks (1332 keV
c 103 •
line from decay o f 6°Co) shown in fig. 1. Only if an
analytical function which adequately represents the
low-energy tailing is used, can the area of a slightly
lower energy peak be determined precisely.
The goal of the work that provided the basis for
this study is to determine 33-ray emission rates with an
accuracy o f 0 . 2 5 - 0 . 5 % . Ideally, this precision would 102

be obtained even for components of closely spaced


but resolved lines. That is, the low-energy tailing
would need to be accounted for.
A variety of analytical functions has been pro-
posed to represent the shape of a 33-ray peak from a
Ge semiconductor detector. Several of these func- 101
3375 34'00 34]25 34150 -- 3475
J 3560 ] 35150 - -~
3575
3525
Channeq
* Work supported by the US Department of Energy under
DOE Contract No. DE-AC07-76ID01570. Fig. la.

499
50O R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks
105 ~ B I I
more free parameters) functions have been indicated.
Also, associated with e a c h p e a k is a step-like increase
in the underlying spectral "background", as indicated
in fig. 1.
In this paper we present a comparison of the qual-
104
ity-of-fit obtained for several previously proposed and
new peak shape functions applied to data for peaks at
1332 keV from three Ge(Li) and Ge detectors. The
better fits are shown to occur for the functions with
the most parameters or the most additive tailing func-
tions. The variation of the computed peak area as a
function of some of the peak shape parameters is also
t~ 103
Z presented. Some aesthetic questions related to the
4 choice of analytical functions to represent the peak
4
4 and the "background" step are discussed.
• P
e• . . • •r~ • v v
2. Experimental data and analysis methods
102
The experimental data for this study were ob-
tained on three detectors: D1, a 65 cm a close-ended
coaxial Ge(Li) which is nine years old; D2, a new 120
cm 3 open-ended coaxial Ge(Li); and D3, a 13 cm 3
•o • ,
• .:"~...-...~... planar Ge. A source o f 6°Co was used and the 1332
...... o,_ --0,--~¢'-':~.,,__ /

101 ~ - . I J I o ~ o ~ o L
,, ..-r
__00~
keV ? - r a y peak was placed in channel 3500 of a
3375 3400 3425 34150 '34175 3500 3525 3550 3575 4096-channel anayzer. The total counting rates were
Channel
less than 2000 per second and about 95 000 counts
105 I I - .....
- T T were collected in the peak channel. The pole-zero
C
adjustment and the base-line restorer on each system
was operated to give the best peak shapes.
The resolutions o f the peaks in fig. 1 are 1 . 7 - 2 . 0
keV, or about 5 channels, with D3 giving the narrow-
est peak. Also D3 has the most symmetric peak. For
104
D1 and D2 the deviation from symmetry is visible at
104 counts, while for D3 it is first visible at 10 a
counts.
The data were analyzed with a general purpose
nonlinear least-squares fitting program. The program
was supplied with about 250 channels of data. Gener-
ally, the parameters of the peak shape function and
two background parameters were varied in the fit.
J d When desired, a parameter could be fixed at its initial
value. The same channel range was used in the fit
tl~ • • • °o )~
with each function and this range was large (about
190 channels). This large range may make the
102 ~ - •

Fig. I. Portions of 7-ray spectra showing the 1332 keV line


from 6°Co on detectors D1, D2 and D3, respectively. The
two curves represent the Gaussian portion from a fit with a
complex peak shape function and an estimated step-function
background. The resolutions of these three peaks are 2.0, 2.0
1013375 3410 34125 34150 34175 3500 ~5~5 '
3550 --3575 and 1.7 keV, respectively.
Channel
R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks 501

reduced-x 2, or X~, values smaller than they would be (i.e., the background radiation); (2) the pulses from
if only the peak area were fitted, but it provided a higher energy 7-rays from the source being measured;
consistent measure of the quality-of-fit to the whole and (3) the pulses from the desired 7-ray, but for
peak region. which enough energy is lost from the sensitive volume
of the detector to put the count in the spectral distri-
bution below the peak. It is assumed that the first
3. A n a l y t i c a l f u n c t i o n s two contributions can be represented by a low order
polynomial (often a linear function). The third con-
3.1. Step-like contribution to background tribution should be represented by a step-like func-
tion. (However, many older analysis programs also
The spectral "background" in the region of a peak, approximated this step-like function by a low order
such as in fig. 1, consists of three components: (1) polynomial. See, for example, the comparison of
the pulses related to radiation from other sources these functions in ref. [10] .)
In this study, the peak o f interest is the highest
Table 1 energy one in the spectrum so, except for a small
Step-like "background" functions suggested in the literature amount of random and coincidence summing, contri-
bution (2) can be neglected. It was also assumed that
Number Date Ref. Function a contribution (1) can be represented by a constant.
Therefore, in this study it was generally assumed that
B1 - - Step 1 for x < X
t0 for x > X the total background could be represented by a con-
stant on each side o f the peak connected by a step-
1 - el (x - X)
B2 1969 [111 2a 2 like decrease near the center o f the peak. A few fits
were carried out to determine the improvement ob-
B3 1970 [12] ~+
1 larctanX-P2 °-x tained when the constants were replaced b y sloping
lines.
B4a .1970 [131 A number of analytical functions which have been
suggested [ 1 1 - 2 0 ] to represent the step-like back-
B4b 1977 [141 b 1-1/ + a -1 ground are listed in table 1. As discussed in the litera-
ture, this step-like increase in the counts probably
1 results from events in which part of the energy of the
B5 1973 [151 1 + e(x-X)/0"75 W 3'-ray escapes from the sensitive volume of the detec-
tor. It is then assumed that if a detector had infinitely
B6a 1974 [161 erf[-(x - X)/P1 ] narrow resolution and no tailing, this would result in
B6b 1976 [17] 1 erfct(x - X)12112o] a step-function pulse distribution (i.e., B1 in table 1)
which cuts off at the 3'-ray energy. If we assume a
B7 1975 I181 (2 - e2(x-x)la) for x < X Gaussian broadening o f the step function, we obtain
½e -2(x-x)lcr for x > X the function B6. Or, if broadened with the experi-
mental lineshape we obtain the function B8. There-
fore, B8 should approach B6 as the low-energy tailing
B8 1978 [191 1 - yi
in the peak is reduced.
1979 [20]
i=x i=O
A comparison of several o f the functions in table 1
a In these functional expressions and those in table 3: x is is shown in fig. 2. The function of entry B8 is not
the channel number or independent variable; Yi is the net included since it depends on the particular peak
count (data minus background) in channel i; X is the center shape; entry B6a would be the same as entry B6b if
of the peak (in channels); Y is the Gaussian amplitude (in normalized and P1 = 21/2a; and entry B4b is similar to
counts); W is the fwhm of peak (in channels); o is the Gaus-
sian half width, 14'= (8 In 2)u2tr; J is the distance (in chan- B4a except for a shift along the channel axis.
nels) from peak center to junction point where Gaussian (Throughout the text, the form used to identify a
function changes to exponential function, J/> 0; P1, P2 are background function is B1, that for a peak shape
the parameters to be determined from fit to experimental function is S1, and that for a parameter in either
data; err is the error function; erfc is the complementary
function is P1.) As indicated, the functions from
error function.
b Kern form as modified by Jorch and Campbell. entries B2 and B4a go to zero at the center of the
502 R.G. Helmer, M.A. Lee /Analytical functions for fitting peaks

,4t T , i
1.2 !~ w =I '

1.0

\%. i

~ 0.6--

0.4-- ~ - -

o B , ~ . . ~- ' - ~ " ~ - - -~'--~


-10 -5 0 5 10
Channel

Fig. 2. Comparison of step-like "baCkground" functions given in table 1 (except entries B4b, B6a and B8). The peak center is at
x=0.

14 b I I

12

1.0

& 0.8
o

~: 0.6

x\

0.4L
0.2

-10 -5 0 5 10
Channel

Fig. 3. Comparison of four step-like background functions from table 1, except the slope (or cut-off rate) determining parameters
of B5 and B7 have been modified to give similar slopes at x = 0.
R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks 503

Table 2 marion o f the counts above the background from X -


Comparison of quality-of-fits with three background func- 1.55W to X + 1.55W (see definitions in table 1), the
tions. Fits are to the peaks shown in fig. 1 and the peaks are area in the low-energy tail is 2.2% o f the total area for
represented by function S15 in table 3
D1 and for the other two detectors this value ranges
Background ×~ down to about 0.8%. Since 98 to 99% of this area is
function in the main peak, any method o f determining this
D1 D2 D3 area which is reproducible should allow area deter-
minations of <1% precision. This assumes the elec-
Linear (no step) 2.97 2.15 2.27 a
tronics and the measurement methods (e.g., count
B1 + constant b 1.92 1.75 2.39
B6b + constant b 1.92 1.75 2.38 rates, source configuration, counting geometry) are
adequate to provide reproducible peak shapes.
a The linear background function has accounted for a small The summation method should be as accurate as
amount of "tailing" on the high-energy side of this peak any other technique with the precision limited by the
and thus produced a smaller x~ than the other functions.
statistics and the ability to define a reproducible
b This constant represents the background on the high-energy
side of the peak. background. This method has the distinct advantage
that the effect of small gain and zero shifts in the
ADC and/or amplifier should be smaller than when
peak. F o r the reason noted above, this is probably fitting with an analytical function. Many methods
not desirable. Since the slopes of the remaining four have been suggested for defining the peak area and
continuous functions, as shown in fig. 2, differ con- background for the summation technique, including
siderably near x = 0, it is difficult to compare the both total peak areas (TPA) and partial peak areas
inherent differences. Therefore, these four functions (PPA). In ref. [10] different backgrounds and chan-
are replotted in fig. 3 with the slope-determining pa- nel ranges in the summation are reported, ref. [21]
rameters of B5 and B7 modified. Here it is seen that, compares a TPA summation to fitting with the
except for the slow approach to the asymptotic val- SAMPO peak shape [5], and refs. [22] and [23] com-
ues by the arctan function, B3, these functions are pare various TPA and PPA methods. Uncertainties of
quite similar. (Function B4b is not shown. It would 0.5% in the 7-emission rate have been reported with
lie among those shown for x < 0 and much below this method and since this uncertainty includes those
them for x > 0, going to zero near x = 2.2.) in defining the efficiency curve the peak areas must
The reduced-x 2 values from fits to the experi- have much smaller uncertainties. The main limitation
mental peaks with one peak shape function were is that only well-resolved single peaks can be ana-
determined for functions B1 and B6b, as well as a lyzed. As noted in the introduction our interest also
linear function. The results, given in table 2, show the includes precise analyses of peaks that are closely
considerable improvement that is obtained by use of spaced.
a stepqike function rather than a linear polynomial.
They also indicate that only a trivially small differ- 3.2.2. Fitting with analytical functions
ence is found between the simple (but discontinuous) A number of analytical functions have been pro-
step function and a continuous function. Even posed to represent the low-energy tailing of the
smaller differences would then be expected between peaks. Thirteen o f these functions are listed in
the various continuous functions shown in fig. 3. table 3, along with two new functions considered in
Due to the small difference in X~t observed for the this study. The functions can be divided into groups
last two cases in table 2, the authors feel that the by how they account for the low-energy tailing on
choice of background step function can be decided the peak. Shapes $2 (Routti and Prussin [5]), $8
on the basis of aesthetics or convenience. (Graham et al. [27]) and S10 (Libert [ 2 9 ] ) m o d i f y
the main Gaussian component and do not have any
3.2. Peak shapes additive tail. Functions $5 (Varnell and Trischuk
[25]), $9 (Baba et al. [28]) and S l l (Schick [30])
3.2.1. Functional fitting or summation modify the main Gaussian shape, but also have an
The contribution of the tailing to the total peak additive tail component. The remaining peaks shapes
area is illustrated in fig. 1 for these detectors at 1332 retain the main Gaussian component and add various
keV. If the area o f the main peak is taken as the sum- numbers of tailing terms. Functions $3 (Sanders and
504 R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks

Table 3
Analytical forms of suggested peak shape functions assuming no high-energy tailing

Number Date Ref. Program Function a


Ram e
Main portion Additive tail

S1 - Many Many Gaussian = Y e - ( x - X ) 2 / 202 None


= y e-i(x-X)/W/(4 In 2) 1/212
$2 1969 [5 ] SAMPO e -(x-X)z/2a2 forx >X J None
{e+J(2x-2X+J)[2°2 for x < X - J

S3a 1969 [ 11 ] C24WMS Gaussian P1 e--P2(x--X) (1 - e +P3(x-X)212a2)


forx <X

S3b 1972 [61 GAMANAL Gaussian Same as 3a

S4a 1969 [24] GSSLRN Gaussian Gaussian(s) - P1 e-(x-X+p2)2/2°2


S4b 1970 [121 Gaussian Same as 4a
S5 1969 [25] See footnote c See footnote c

$6 1970 [131 Gaussian

& - x~'27
/

X P3 eP4(x-X)]° for x < X

S7 1971 [26] Gaussian P1 e P2(x- X)? /1 1 ~ P 3 ~x ~ ~ ~

$8 1972 [271 PI f e-P2(x-t)2 e-P3(x-t) dt Included in main portion

$9 1973 [28] Asymmetric Gaussian Same form as main peak with same widths

e-(x-X) /2or 2, x > X

e -(x X)2/2PI2, x < X

S10 1973 [29] Gaussian except

o = Pl (1 + (P-3-22- 1)(x - Included in main portion

Sll 1974 [30] SKEWGAUSS Same as $2 eJ(2x 2X+J)/2°2 Pl(X - X + j)n


with n = 4 suggested

$12 1976 [171 HYPERMET Gaussian b 0.5 P1 e(x-X)/P2 erfc x - X + x/2aPs 1


(x/2~P5 2P2 ]
(x-X x/2oP61
+ 0.5 P3 e(x-X)/P4 erfc x/2aP6 + 2P 4 ]

S13 1979 [91 Gaussian P1 e-(x-P2)2 /2P32 , x > P2 2P32


P4

P1 e-(X P2)/P4, x < P2 - 2P3~2


P4
R.G. Helmet,M.A. Lee/Analytical functionsfor fitting peaks 505

Table 3 (continued)

Number Date Ref. Program Function a


name
Main portion Additive tail

$14 1979 This report Gaussian P1 e-(x-X+P2)212P32+P4 e(x-X)/Ps


XerfcI~ + ~5 ]X2/°P6~

$15 1979 This report Gaussian P1 e-P2(x-X) (1 - e-P3(x-X)2/2°2)


+P, e(x-X)/Ps e r f c ( ~ +~ P6)

a See table 1 for definitions of the various symbols and functions.


b See text for discussions of interaction between Gaussian and short-term tailing.
c In ref. [25] the main peak is represented by
(x-X) 02 x-X o2
P1 2(P2 + P3) e P2 1-
erf(X + o
\ x/2o x/2P2
+ e P3 2P3 1-

which probably includes the short-term tailing and the long-term tailing is represented by
erf -(xx/2o
°1}
+ x/2p 3

x-X 02
P4
2ps
e
P5 2P52
- erf
x- X
-- +
o
x/2o x/2es
)].

Holm [11] or Gunnink and Niday [6]), S4b (Robin- would not decrease monotonically below the peak.
son [12]), $7 (Hoclet [26]) and S13 (Campbell and This is the same disadvantage as shown by shape S 11
Jorch [9]) have one additive tail component; S12 as discussed in the next section. The functions of
(Phillips and Marlow [17]), S14 and S15 (suggested Hoclet [26], $7, and Baba et al. [28], $9, were also
in this report) have two additive tail components; and not considered.
$6 (Kern [13]) has three additive tail components. Table 4 gives the reduced-x 2 values, X~t, from our
The function S14 combines a Gaussian additive tall fits along with the number of additive tailing compo-
term with an exponential term of the form used in nents in each function and the number of parameters
HYPERMET [ 17] ; while S 15 combines two exponen- (not including the parameters in the step-function
tial terms, one of the GAMANAL form [6,11] and background). One should not overemphasize the signif-
one of the HYPERMET form [17]. icance of small differences in ×2- Due to the use of a
Nine of these functions have been fitted to the discontinuous search process for some parameters
three measured 1332 keV pealts in fig. 1. These nine (e.g., P3 in function $3), the exact minimum may not
include those of several well-known programs (e.g., have been found. Also, there is some freedom in the
SAMPO and GAMANAL). The peak shapes we have number of peak-shape parameters; for example,
not considered in this study include those of Varnell HYPERMET can be used with 7, 8 or 9 variable
and Trischuk [25], S5, Graham et al. [27], $8, and parameters.
Libert [29], S10. These functions include some, or The results of a number of these fits are also illus-
all, of the tailing in the main peak function and at trated in figs. 4 - 7 . These show the deviations
least that of Graham et al. [27] may be of interest between the experimental and analytical functions
because it derives the analytical function on the basis for detector D1 for shapes with 0, 1 and 2 additive
of arguments as to how a detector works. The func- tailing components (3, 4, 6 and 8 parameters). It is
tion of Kern [13], $6, is quite complex and from its clear from fig. 7 that for the 8-parameter function, it
form we assume it could result in a shape which is reasonable to consider the low-energy asymmetry
506 R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks

Table 4
Comparison of reduced-x 2 values for fits with several ana-
lytical functions. The experimental data are shown in fig. 1.
The step-like background component has zero slope on each
1°5I - - - - - - r . . . . -T" . . . . ~*,

side of the peak


104
Addi- Peak Peak Reduced-× 2 values
tire shape shape
tail param- function D1 D2 D3
corn- eters I
po-
nents 103 ~

0 3 $1 60.0 49.9 13.2


L_ ............ •, . . . . . . mz~gLZ~ .....
0 4 $2 40.1 26.5 (13.5) a
L bjg~l~u
1 6 $4 c 14.2 7.26 4.21 I
1 6 Sll b 6.58 4.32 4.47 102
1 6 $3 8.15 3.15 3.04
1 7 S13 6.57 2.52 2.76
2 8 S14 c 2.36 2.15 2.99
.... ) -*- yJ
2 8 S15 e 1.78 1.75 2.39 ,o 4e* % "*'
2 8 S12 d 1.54 1.77 2.36
101 I I __._2

a This value assumes the exponential junction point is near --!


# •
the peak. If the junction point becomes far from the peak,
10 f~'~
the Gaussian value must be obtained. Ob
b One parameter is an integer power of a polynomial term
which is fixed in the fitting process.
lO
c In these fits the widths of the two Gaussians are indepen-
dent. i
~°3350 34'06 ~3450 3500
"" 3500 3600
d In these fits Ps = P6 so function has eight parameters. Channel
e In these fits P6 = 1.0 so function has eight parameters.
Fig. 4. Results of the fit to the 1332 keV peak from detector
D1 with a Gaussian function, S1, and step-function back-
ground, B1. The solid curve represents the total analytical
as having a short-term (i.e., within the " p e a k " ) and a function and the dashed curve represents the step back-
long-term (i.e., extending well below the peak) com- ground. The points at the bottom give the difference between
the data and the analytical function in units of the square
ponent.
root of the counts.
The results in table 4 indicate that the more com-
plex functions generally give the b e t t e r fit. However,
for some detectors the functions w i t h one additive
tall c o m p o n e n t (i.e., the 6-parameter functions) m a y value o f P3 is about the same for all three detectors
give suitable fits (X~ ~ 3) if one prefers n o t to use the and that P3 can be varied a factor of about two w i t h
more c o m p l e x functions. only a small change in X~. A similar result w o u l d be
The level o f difficulty in the use of the m o r e com- e x p e c t e d for the cut-off parameters Ps and P6 in
plex o f these functions (especially S12 and S 1 5 ) w i l l function S12. In fact, Phillips [31] suggests setting
depend in part on h o w easy it is to determine the var- Ps = P6 = 1.0 in all cases.
ious parameters. The first examples to consider are Similarly, the rate o f rise o f the " H Y P E R M E T -
the parameters (P3 in $3 and S15 and Ps and P6 in t y p e " long-term tailing c o m p o n e n t (/°4 in S12 or Ps
S12) which determine the rate of decrease (or cut- in S15) m a y be a rather insensitive parameter. This
off) of the tailing terms near the peak center. For the possibility is indicated by the results in fig. 9 which
detectors discussed by G u n n i n k and Niday [6], t h e y indicate that for detectors D1 and D2 a wide range of
were able to use a value P3 = 0.4 in f u n c t i o n $3 for all Ps values give similar X~ values.
detectors at all energies. Fig. 8 shows the reduced-x~ F r o m the results presented in figs. 8 and 9, it is
values for fits to the three peaks w i t h functions $3 clear that several of the parameters in functions $3,
and S15 and P3 fixed at the values shown. These S12 and S15 can often be fixed at p r e d e t e r m i n e d val-
results show that at this 3,-ray energy the o p t i m u m ues and, thereby, the n u m b e r o f variable parameters
R.G. Helmer, M.A. Lee/Analytical functions for fitting peaks 507

105 105
tl

I
!
lo4 ~- 104 --

~ 103
L)

~+• ,~++ t.+


~k~+"+;~+""" ~' t
t
102 - - I
I
I
. . . . .
~..*l e4,- - --- e~ i'D);'b, o . ~•- -
l
101 L ~ I L __ I . . . . j

20f
10

1°0 .+,~,~,+,%,.~,+,;,;.+.:,.
e,,•e -


. •
e•
0++,'++,,'.+~*+:,.++,,.,
• o • v •
-10 L
o•
-20 _ _ _ _
3350 34100 34J50 +3500 35150 3600 -10 i
3350 34100 3450- 35100 3550 3600
Channel
Channel

Fig. 5. Results of the fit to the 1332 keV peak from detector Fig. 6. Results of the fit to the 1332 keV peak from detector
D1 with SAMPO function, $2, and step background, BI. The D1 with the GAMANAL-type function, $3 and step back-
form is as in fig. 4, except the long-dashes represent the ground, BI. The form is as in fig. 4 except the long dashes
exponential part of the peak shape function. represent the additive tailing term.

reduced. This may make these functions more man- was obtained with n = 6 or 8 which represents an
ageable. intermediate situation. With these values, this expres-
F o r the fits reported in table 4 and figs. 4 - 7 , it sion produces a hump in the total analytical function
has been assumed that there is no high-energy tailing between 10 and 15 channels below the main peak
and that the background above the peak is a constant. center (i.e., the function did not decrease mon0toni-
However, fig. 1 indicates some slope to this back- cally below the peak). Therefore, this function is not
ground. (This slope may result in part from sum- considered useful when used in this way, especially
ming.) Therefore, as given in table 5, a few o f the bet- when compared with the other functions with one
ter fits were repeated with a sloping background. The additive tail component.
same slope was required on each side of the peak. The The function proposed by Campbell and Jorch
improvement in the fits is only significant for the D3 [9], S13, has one additive tail component and
detector. involves 7 parameters. The quality-of-fits obtained
were intermediate between those o f the 6- and 8-pa-
3.2.3. A e s t h e t i c considerations rameter functions. These authors suggest that the
The single tailing term of the SKEWGAUSS func- parameters of their function can be estimated more
tion [30], S l l , can be used to represent short- or easily than those o f the GAMANAL type tailing func-
long-term tailing by changing the power n of the tion, $3. This fact, their sequential method of param-
polynomical tenn. In our tests the minimum ×~t value eter estimation, and the determination o f the width
508 R.G. Helmer, M.A. Lee / A nalytical functions for fitting peaks

10 5 r i 11 I r I I I I

",,,,.
\

10
I
\
104 9

8 D1, S3

ii

o 103 tl

-'.'a.:~ ~;,,+'~+'.+:';~i~'`,'`':'+: " "---/1-


102 I !
// iT~~,,.'..,... ... . . . \

101 i I I --*--6--|--1--1---------'$__=__ = _

D3, S3
D3, $15
6--6--6--A--6--A--A

l~l D
2
0
; •
"11
• o.~*
oe_"
0.0
o o • o~
l,

o**
"o" • • •
°o% •
..o•,~,o.._
lo •
"
4
I %"e ~ e. - - ' ' e
._~ o/eD 1, $15
.--"
~j
_
~,~_e,
"...q'~.,'~,***.t
-
%,t,'~% g "+~# • "• I
D2, $15

+ -:"•" .. -": • " ....1- "o+


' 0!2 0!4 0!6 0!8 ,!0 ,!2 14
J I I ___ P3 (channels)
43350 3400 3450 T ~ e [3500 3550 3600
Channel
Fig. 8. Reduced-× 2 values for fits to the 1332 keV peaks
Fig. 7. Results of the fit to the 1332 keY peak from detector from detectors D1, D2 and D3 with peak shape functions $3
D1 with the GAMANAL-type short-term tail and and S15 for various values of the cut-off parameter P3. A
HYPERMET-type long-term tail, S15, and erfc-type step step-like background, B1, was used.
background, B6b. The form is as in fig. 4, except both addi-
tive tailing terms are shown.

W only from the Gaussian term are all ideas from ref. fit with function $3. If we include the area of the
[9] with significant merit. Campbell and Jorch [9] short-term taft in the peak area, this should not cause
developed this function for use at low energy and any problem in obtaining precise answers. In fact, for
illustrate it with fits at 84 keV. a detector with a large amount of tailing, this large
Results obtained with the last two 8-parameter percentage loss from the Gaussian area may be realis-
peak shape functions in table 4, $12 and S15, are tic. That is, the short-term tailing could include
quite comparable. Although they give ,fits that should events which are still in the main peak, but which
be quite acceptable, each o f these functions has a fea- have been displaced to a lower energy.
ture which a particular user may fred aesthetically The tailing function of $3 (GAMANAL) has the
objectionable. In HYPERMET, S12, each o f the tail- feature that it goes to zero at the center of the peak.
ing terms is decreased to about one-half of its maxi- Conversely to the HYPERMET type term, this is good
m u m value at the center o f the peak. (As discussed in in that this term does not interfere with the Gaussian
ref. [17], this is reasonable if we consider them to area, but it may be considered aesthetically poor
result from the broadening of a function which, for since the physical origin o f these events would imply
infinitely good resolution, would go to zero at the some contribution above the peak center.
"r-ray energy.) As a result, the short-term tail compo- F o r whatever tailing function is used, we must
nent can "absorb" a significant fraction o f the "Gaus- decide on a definition o f the peak area. For the 8-pa-
sian area". In section 3.3 a case is noted where the rameter functions (see fig. 7), it seems quite reason-
Gaussian area is 30% less than the Gaussian area for a able to include the short-term tail component in the
R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks 509

105

104

1i
103

102 // I' ',~J .

/
/
/
/
101 I I / L
l
4 I I °--/'~ o• ....~ -

2
• • oo • o• so

1 5I 110 115 /0 15 30 35
.No•
P5 (channelsl 2

Fig. 9. Reduced-x 2 values for fits to the 1332 keY peaks -4


iI
3350 34100 T 34150 3500 3510 3600
from detectors D1, D2 and D3 with peak shape function S15 Channel
for various values of the parameter P5 which determines the
rate of rise of the long-term tailing component. Fig. 10. Results of the fit to the 1332 keV peak from detec-
tor D3 with the GAMANAL-type function, $3, and step
background, B1. Form is as in fig. 4.
area a n d t o e x c l u d e t h e l o n g - t e r m c o m p o n e n t . I n con-
trast, for t h e 6 - p a r a m e t e r f u n c t i o n s , t h e decision m a y
d e p e n d o n the d e t e c t o r . In fig. 6, it a p p e a r s reason-
able t o i n c l u d e t h e tailing as p a r t o f t h e p e a k area.
B u t for a d e t e c t o r w i t h a " b e t t e r " p e a k shape, as Table 6
s h o w n in fig. 10, this single tailing f u n c t i o n i n c l u d e s Peak areas for various fitting functions. The spectrum is from
c o n t r i b u t i o n s well b e l o w t h e peak a n d it is n o t clear detector D1 and background B1 or B6 was used

Peak Area
shape
Table 5 function Main Tail Total
Comparison of fits with zero-slope and sloping step-like Gaussian component
backgrounds. The fits with the zero-slope backgrounds are
the same as those in table 4 $1 549400 - 549400
$2 400600 a 154300 554 900
Peak shape Back- x~ $4 515 300 44 100 559 400
function ground $3 516 100 44600 560700
slope D1 D2 D3 S13 494600 68300 561 900
S14 473400 75400 548800
S15 0 1.78 1.75 2.39 S15 497 800 56 900 554 700
~0 1.76 1.62 1.96 S12 355 800 197000 552700
S13 0 1.54 1.77 2.36
~0 1.54 1.65 1.96 a This includes only the Gaussian area above the junction
point with the exponential tall.
510 R.G. Helmer, M.A. Lee / AnalytT"calfunctions for fitting peaks

Table 7
Variation o f peak area with value of cut-off and exponential slope parameters. Data are for the functions and parameter ranges
shown in figs. 8 and 9

Peak Detector Fixed Area Variation a


shape parameter (%)
Main Tall Total
Gaussian component

$3 D1 P3 = 0.2 522 970 37 600 560 570 -0.11


0.4 523 080 37 850 560 940 -0.04
0.6 520 090 40 370 561 230 +0.01
0.8 517 480 43 600 561 080 -0.01
1.0 514 780 46 380 561 160 -0.00
D2 P3 = 0.2 493 670 24 390 518 060 -0.12
0.4 492 380 26 050 518 430 0.05
0.6 490 080 28 390 518 470 -0.04
0.8 487 730 30 930 518 650 0.00
1.0 485 570 33 100 518670 --0.00
D3 P3 = 0.2 469 620 5240 474 860 -0.08
0.4 469 250 5890 475 140 -0.02
0.6 468 910 6290 475 200 0.01
0.8 468 650 6580 475 230 =0.00
1.0 468450 6780 475 230 0.00
1.2 468 290 6940 475 230 0.00
S15 D1 P3 = 0.4 498 180 56110 554 300 -=0.00
0.6 497 180 56 820 555 050 +0.14
0.8 496 050 59 660 555 710 +0.25
D2 P3 = 0.4 482410 29830 512250 0.28
0.6 481 550 32 160 513 710 =0.00
0.8 497 620 34 840 514 460 +0.15
D3 P3 = 0.4 461 360 7970 469340 0.02
0.6 461 200 8240 469 440 -=0.00
0.8 461 040 8510 469 550 +0.02
S15 D1 P5 = 5 484 963 49 027 533 990 3.7
10 491 128 57 752 548 880 -0.99
15 494 786 57 763 522 549 -0.32
20 497 343 57 003 554 346 ~0.00
25 499 028 56 389 555 417 +0.19
30 500 207 55 899 556 106 +0.32
D2 Ps= 5 467 183 25592 492775 4.7
10 478 153 32 563 510716 1.2
15 480 319 36 561 516 880 =0.00
20 481 311 33 923 515 234 -0.32
25 481 924 33 966 515 890 -0.19
30 482 351 31 869 514 220 -0.51
D3 Ps= 7.5 460 352 6609 466 961 0.62
10 460 119 8485 468 604 0.27
15 461 872 8016 469 888 ~0.00
20 463 773 7033 470 006 +0.02
25 464 899 6569 47l 468 +0.34
30 465 592 6348 47l 940 +0.44

a The reference value is taken to be that where x ~ is a m i n i m u m .


R.G. Helmet, M.A. Lee ~Analytical functions for fitting peaks 511

whether the tailing should be included in the peak three detectors, the maximum difference is 0.12% as
area. P3 (cut-off rate) varies by a factor of 5 (from 0.2 to
1.0). This consistency may make this (or a similar)
3.3. Peak areas function of interest for precise measurements and
supports the idea that this parameter could be fixed
Although the peak areas from the fits carried out at a predetermined value in most cases.
in this study do not have any special meaning, some For shape S15, which has two tailing terms the
qualitative ideas can be obtained from them. [The first of which is identical to that of $3, only the
areas of the Gaussian components (except that from short-term tail is included in the peak area. The sec-
$2, SAMPO) are the analytical integral, i.e., ond section of table 7 indicates that for S15 the total
(n/4 In 2) 1/2 YW, and for the exponential compo- peak area varies up to 0.43% as Pa varies over the
nents the numerical integral (summed over all chan- more limited range of 0.4 to 0.8. It is presumed that
nels with counts greater than 1.0) was used.] All fits this larger variation (eight times as large as with $3
were with the background functions B1 or B6. over this range) is due to the changes in the amount
The "peak areas" for the spectra on detector D1 of the tailing that is taken up by the long-term tail
are listed in table 6 for a number of the fits whose X~ component. It is assumed that any function with
values are given in table 4, including those for the fits similar tailing terms (e.g., S12 or S14) would have the
shown in figs. 4 - 7 . The total area has a range of 2.2% same behavior.
(from 5 . 4 9 × 1 0 s - 5 . 6 2 × 10s), and the variation The third section of table 7 shows the variation in
occurs in an understandable manner. As indicated in the total area when the slope, Ps, of the long-term
fig. 4, the Gaussian function, S1, gives a relatively low exponential tail of S 15 is varied. Here the changes are
area. The functions $2, $4, $3, and S 13, with a single 0.7-1.7% for changes of Ps from 10 to 30 channels
tailing term, have forms which include successively (for W ~5 channels).
more of the tailing and therefore have larger areas. In Finally, it is satisfying to note that the detector
these cases all of the tailing function is included in with the best peak shape, D3, also has the smallest
the total peak area. For the functions with two addi- variation in the computed peak area in all three sec-
tive tailing functions, the long-term component is not tions of table 7. This difference is especially note-
included in the peak area even though it extends to worthy in the second section of the table where the
the center of the peak, so these peak areas will be variation is a factor of 6 and 11 smaller than for the
smaller. This is indicated for functions S12, S14 and other detectors. This suggests that for precise spectro-
S15 in table 6. The Gaussian tail of S14 does not scopy there may be an advantage in using a detector
include as much of the tailing as the exponential with a good peak shape.
short-term tails of S 12 and S 15, so this function gives
the smallest total area. As noted earlier, S12 allows a
large transfer of counts from the Gaussian to the tail- 4. Discussion
ing component.
For all three detectors, the total peak areas have The results of this study give a general idea of the
also been determined as a function of the parameters complexity needed in an analytical function in order
defining the rate of cut-off of the tailing (Pa in $3 to fit the shape of an experimental ")'-ray peak from a
and S15) under the peak and the slope of the long- Ge or Ge(Li) detector. Anyone designing a new ana-
term exponential tail (Ps in S15). The results are lysis program can weigh the benefits of the better fits
given in table 7, where the variation in the last col- against the disadvantages of the many parameters
umn indicates the "error" that might be produced by involved in the more complex functions. For precise
fixing the parameter since the variations are refer- area determinations, the sensitivity of the computed
enced to a fit giving the minimum X~ value. These fits area to the shape parameters may need to be investi-
are the ones used to generate figs. 8 and 9. Generally, gated for a particular detector before a final choice is
the regions with significant increases in X~ are those made.
of the largest changes in the computed area. The results given here constitute only the begin-
An interesting point from table 7 is the extreme ning of a study of )'-ray peak shapes. The next step
consistency shown in the first part of the areas for would be to determine the parameters in the peak
peak shape $3 which has a single tailing term. For all shape function (except Y and X) at various )'-ray
512 R.G. Helmer, M.A. Lee / Analytical functions for fitting peaks

energies. Since in the analysis of most spectra these [5] J.T. Routti and S.G. Prussin, Nucl. Instr. and Meth. 72
shape parameters need to be fixed at known values (1969) 125.
(i.e., the data are usually not of sufficient quality to [6] R. Gunnink and J.B. Niday, University of California
Radiation Laboratory Report, UCRL-51061, vol. 1 4
allow ther.' ~o be determined), some simple analytical
(1972).
functions are needed to represent the parameter val- [7] C.M. Lederer, University of California Radiation Labo-
ues as a function of the energy. ratory Report UCRL-18948 (1969).
The prospective user must also determine how [8] L.A. McNelles and J.L. Campbell, Nucl. Instr. and
these parameters are affected by the counting rate. Meth. 127 (1975) 73.
[9] J.L. Campbell and H.H. Jorch, Nucl. Instr. and Meth.
This intbrmation can then be used to limit the count- 159 (1979) 163.
ing rate to a range that is compatible with the known [10] R.G. Helmer, J.E. Cline and R.C. Greenwood in: the
parameter functions, or to adjust the parameters to Electromagnetic Interaction in Nuclear Spectroscopy,
the counting rate. ed., W.D. Hamilton (North-Holland, Amsterdam, 1975)
p. 775.
In all of the results quoted herein, the ~(~ values
[11] W.M. Sanders and D.M. Holm, Los Alamos Scientific
are based on a(y)=yUZ where y is the count in a Laboratory Report LA-4030 (1969).
channel. Since the maximum count, about 10 s, is not [12] D.C. Robinson, Nucl. Instr. and Meth. 78 (1970) 120;
extremely large this may be fairly realistic. However, Report AERE-R 6144 (1969).
deviations from this relation might be expected on [13] J. Kern, Nucl. Instr. and Meth. 79 (1970) 233.
the sides of the peak where the effect of a gain or [14] H.H. Jorch and J.L. Campbell, Nucl. Instr. and Meth.
143 (1977) 551.
zero shift would be the largest. As a result it may be [15] W. Teoh, Nucl. Instr. and Meth. 109 (1973) 509.
unrealistic to except X~ values of about 1.0 unless we [16] M. Dojo, Nucl. Instr. and Meth. 115 (1974) 425.
increase the uncertainties in these regions. [17] G.W. Phillips and K.W. Marlow, Nucl. Instr. and Meth.
For "small" computers, the complex functions 137 (1976) 525.
discussed here may appear formidable. However, it [18] N. Sasamoto, K. Koyama and S. Tanaka, Nucl. Instr.
and Meth. 125 (1975) 507.
should be noted that the step-like background B8
[ 19] R. Gunnink, in: Computers in Activation Analysis and
(which involves primarily additions) is especially use- Gamma-Ray Spectrs, Proc. Conf. at Mayaguez, Puerto
ful for small computers. Also, Phillips [32] has sug- Rico 1978 (1979) p. 109; Report CONF-780421.
gested a fast method for approximating the comple- [2o1 W. WStzig, in: Computers in Activation Analysis and
mentary error function used in S12 and S15 which Gamma-Ray Spectroscopy, Proc. Conf. at Mayaguez,
Puerto Rico 1978 (1979) p. 150; Report CONF-
should make them manageable on small computers.
780421.
[21] H.P. Yule and H.L. Rook, J. Radioanal. Chem. 39
The authors wish to thank R.L. Heath for his con- (1977) 255.
tinuing support of developmental work in 7-ray spec- [22] L. Kokta, Nucl. Instr. and Meth. 112 (1973) 245.
[23] J. Hertogen, J. DeDonder and R. Gijbels, Nucl. Instr.
troscopy and M.H. Putnam and E.W. Killian for their
and Meth. 115 (1974) 197.
programming support for spectra analysis. The I241 G.D. Seybold, Battelle Memorial Institute Report
authors also acknowledge conversations with and sug- BNWL-1227 (1969).
gestions of G.W. Phillips and R. Gunnink. [251 L. Varnell and J. Trischuk, Nucl. Instr. and Meth. 76
(1969) 109.
References [261 M. Hoclet, Commission of Atomic Energy Report CEA-
4199 (1971).
[27] R.L. Graham, J.S. Geiger and M.W. Johns, Can. J. Phys.
[1] K. Debertin, U. Sch6tzig, K.F. Walz and H.M. Weiss, 50 (1972) 513.
Proc. ERDA Symp. on X- and Gamma-Ray Sources and [281 H. Baba, T. Sekine, S. Baba and H. Okashita, Report
Applications, CONF-760539 (1976) 59. JAERI-1227 (1973).
[2] A.F. Hirshfeld, D.D. Hoppes and F.J. Schima, Proc. [291 J. Libert, Nucl. Instr. and Meth. 109 (1973) 609.
ERDA Syrup. on X- and Gamma-Ray Sources and Ap- [301 W.C. Schick, Jr., Ames Laboratory Report IS-3460
plications, CONF-760539 (1976) 90. (1974).
[3] Y. Yoshizawa, Y. Iwata, R. Kaku, T. Katoh, J. Ruan, [311 G.W. Phillips, in: Computers in Activation Analysis and
T. Kojima and Y. Kawada, Hiroshima University Report Gamma-Ray Spectroscopy, Proc. Conf. at Mayaguez,
HUPD-7919 (1979). Puerto Rico 1978 (1979) p. 197; Report CONF-
[4] R.J. Gehrke, R.G. Helmer and R.C. Greenwood, Nucl. 780421.
Instr. and Meth. 147 (1977) 405. [321 G,W. Phillips, NRL Memorandum Report 3963.

You might also like