IUPAC

Guidelines for the use of atomic weights

Fo
Journal: Pure and Applied Chemistry

Manuscript ID PAC-REC-16-04-01
rP
Manuscript Type: Recommendation

Date Submitted by the Author: 01-Apr-2016

Complete List of Authors: van der Veen, Adriaan; VSL

ee

Meija, Juris
Possolo, Antonio; National Institute of Standards and Technology
Hibbert, David; University of New South Wales, School of Chemistry
rR

atomic weights, atomic-weight intervals, molecular weight, standard

Keywords:
atomic weight, measurement uncertainty, uncertainty propagation

Author-Supplied Keywords:
ev
ie
w
On
ly

P.O. 13757, Research Triangle Park, NC (919) 485-8700

Page 1 of 13 IUPAC

Pure Appl. Chem. 2016; aop

1
2
3
Sponsoring body: IUPAC Inorganic Chemistry Division Committee: see more details on page XXX.
4
5 IUPAC Recommendation
6
7 Adriaan M. H. van der Veen*, Juris Meija, Antonio Possolo, and D. Brynn Hibbert
8
9
10
Guidelines for the use of atomic weights 5

11
12 DOI: ..., Received ...; accepted ...
13
14 Abstract: Standard atomic weights are widely used in science, yet the uncertainties associated with these
Fo
15 values are not well-understood. This recommendation provides guidance on the use of standard atomic
16 weights and their uncertainties. Furthermore, methods are provided for calculating standard uncertainties
17 of molecular weights of substances. Methods are also outlined to compute material-specific atomic weights 10
18 whose associated uncertainty may be smaller than the uncertainty associated with the standard atomic
rP
19
weights.
20
21 Keywords: atomic weights; atomic-weight intervals; molecular weight; standard atomic weight; uncertainty;
22 uncertainty propagation
ee

23
24
25 1 Introduction 15
26
rR

27 Atomic weights provide a practical link the SI base units kilogram and mole. For every substance, or
28
mixture of substances, the conversion from mass to amount of substance (chemical amount (6)) and vice
29
versa can be carried out once the molecular weight has been determined. Standard atomic weights play a
30
31 key role in these calculations. These values are regularly evaluated and published by the Commission on
ev

32 Isotopic Abundances and Atomic Weights (CIAAW, www.ciaaw.org) of the International Union of Pure 20
33 and Applied Chemistry (IUPAC). The work undertaken by CIAAW to keep the standard atomic weights
34 up-to-date deals broadly with two aspects: appreciation of improved measurements of isotopic composition
ie

35 of the elements, and the improved knowledge about natural variations of the isotopic composition of the
36 elements (5). On the advice of CIAAW, in 2009 IUPAC resolved to disseminate standard atomic weights
37 25
w

of elements using either of two different notations (17):

38
39 1. traditional short-hand notation giving a single value followed by an expanded uncertainty between
40
parenthesis applicable to the last quoted digit of that value, e.g., Ar (Cd) = 112.414(4), and
On

41
42 2. interval notation for elements whose isotopic composition varies appreciably in nature, e.g., Ar (C) =
43 [12.0096, 12.0116] representing the interval of atomic-weight values of these elements found in “normal”
44 materials (11). 30
45
ly

46
47 The expanded uncertainty used in the traditional short-hand notation is to be understood so that for
48 example for cadmium, the atomic weight of cadmium in any “normal” material is practically certain to lie
49 inside the interval [112.410, 112.418]. While the new interval notation did emphasize the magnitude of the
50 natural variations of atomic-weight values, for many users it was unclear how to employ the given intervals
51
in calculations and uncertainty evaluations. 35
52
53 These Guidelines provide a basis for understanding the standard atomic weights and their uncertainties
54 so that users can readily employ them in a variety of calculations. Informative examples are provided to
55
56 *Corresponding Author: Adriaan M. H. van der Veen: Van Swinden Laboratorium (VSL), Delft, The Netherlands
57 Juris Meija: National Research Council Canada (NRC-CNRC), Ottawa, ON, Canada
58 Antonio Possolo: National Institute of Standards and Technology (NIST), Gaithersburg, MD, United States of America
59 D. Brynn Hibbert: UNSW, Syndney, Australia
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 2 of 13

2 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
3 illustrate the computations and to enable their implementation in practice, in software, or in quality control
4 documents, among others. These Guidelines are accompanied by an IUPAC Technical Report (12) that
5 provides a more in-depth treatment of the subject. The authors wish to emphasize that such calculations
6 should always be performed with the most recent values for the (standard) atomic weights. Throughout
7 5 these Guidelines the standard atomic weights of 2013 are used (9).
8
9
10
11 2 Terms and definitions
12
13 1. atomic weight
14
relative atomic mass, Ar (E)
Fo
15
Amount-weighted average of the atomic masses of the isotopes of an element E in a material, divided
16
17 10 by the unified atomic mass unit (dalton).
18 2. normal material
rP
19 Material that is a reasonably possible source for an element or its compounds in commerce, for industry
20 or science, and the material is not itself studied for some extraordinary anomaly, and its isotopic
21 composition has not been modified significantly in a geologically brief period (11).
22 15 3. standard atomic weight
ee

23 CIAAW-recommended atomic weight, applicable to all normal materials.

24
4. molecular weight
25
26 Sum of the atomic weights of elements in a substance, weighted by their stoichiometric coefficients.
rR

27 5. standard molecular weight

28 20 Molecular weight calculated using the standard atomic weights of its constituent elements.
29 Note: Standard atomic weights are commonly used for calculating molecular weights of normal mate-
30 rials.
31
ev

6. reference atomic weight

32 CIAAW-recommended atomic weights applicable to materials from specific sources.
33 25 Note: In CIAAW reports, reference atomic weights are given for elements whose standard atomic weight
34
is given in interval notation for specific sources such as air, water, or rocks (9).
ie

35
36 7. delta-zero reference material
37 Reference material that is well-characterised used to establish the origin of a delta scale.
w

38
39
40
3 Standard atomic weights
On

41
42
43 30 Many forms of notation have been employed over the years to express standard atomic weights (1; 14).
44 CIAAW currently employs, as mentioned previously, two notations: the single-value shorthand notation
45 with the explicit uncertainty given in parenthesis and the two-value interval notation with no uncertainty
ly

46 given explicitly. Thus, an atomic weight interval is characterised by a lower limit, a, and an upper limit, b,
47 whereas the traditional short-hand notation of the standard atomic weight provides a single value and the
48
35 associated expanded uncertainty (Table 1). It is important to stress that neither of these notations alters
49
the meaning of the quantity, and that both summarize CIAAW decision to express the standard atomic
50
51 weight of an element using two values; either with an upper- and lower-bound or with the representative
52 value and its uncertainty. For both notations, the assurance is given that the atomic-weight values of
53 elements in normal materials are expected to lie between the end-points of the interval given explicitly, or
54 40 within the interval corresponding to the representative value plus/minus the quoted uncertainty (17).
55 Standard atomic weights and their uncertainties are determined by CIAAW (17; 11). In the process
56 leading to determinations of these standard atomic weights, all published data concerning the isotopic
57 compositions of “normal materials” for each element are evaluated, and a value with an indication of the
58
uncertainty, or an atomic-weight interval is established. In this process, the uncertainty is attributable to
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 3 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 3

1
2
3 both variations in the isotopic composition of the element in normal materials, and the uncertainty about
4 the masses of the isotopes (17; 11).
5 Unless interval arithmetic is employed, the majority of scientific calculations require a single, repre-
6 sentative value of the quantity and its uncertainty. Extracting a representative value and an indication of
7 associated uncertainty from a standard atomic weight involves assignment of a probability distribution that 5
8 explicitly or implicitly represents the state-of-knowledge of the standard atomic weight in the material(s)
9
of interest.
10
The representative value may be the expected value (mathematical expectation, or mean) of that distri-
11
12 bution, and the indication of associated uncertainty may be the corresponding standard deviation (standard
13 uncertainty) (2). However, and most importantly, the expected value does not necessarily represent the 10
14 “most-likely” value (mode) of that probability distribution, which is the value where the corresponding
Fo
15 probability density function achieves a maximum. In general, the mode and the expected value may differ.
16 Using CIAAW notations for the standard atomic weight of an imaginary element, [99, 101] or 100(1), nei-
17 ther would imply that a value of 100 is more likely to occur than any other value between 99 and 101. Thus,
18 both notations tell us that the atomic weight in any normal material will be greater than or equal to 99 15
rP
19
and will be less than or equal to 101. The distinction between the two notations serves to emphasize that
20
for some elements (e.g., hydrogen or carbon) the atomic-weight values vary significantly in nature between
21
22 materials whereas for other elements (e.g., fluorine or zinc) our knowledge about the atomic weights is
limited mainly by the uncertainty about the nuclide masses and their abundances.
ee

23
24
25
26
4 Probabilistic interpretation of standard atomic weights
rR

27 20
28
29 The concept of measurement uncertainty applies not only to quantities whose values are determined ex-
30 perimentally, but also to all derivative quantities whose calculation involves such experimental data. The
31
ev

Guide to the expression of uncertainty in measurement (GUM) (2) and its supplements (3; 4) provide
32 a harmonized basis for evaluating measurement uncertainty. The concept of “measurement uncertainty”
33 conveys the notion that a margin of doubt about the true value of a quantity remains after measurement, 25
34
and generally recognizes that the knowledge about this true value is imperfect or incomplete. Thus, if
ie

35
36 an isotopic composition of an element in a material has not been directly determined by measurement of
37 the material, information about the material still may enable the assignment of a value to the isotopic
w

38 composition along with an associated standard uncertainty for use in calculations. Such assessment leads
39 to larger uncertainties than in situations in which accurate isotopic composition data are at hand. 30
40 Application of the law of propagation of uncertainty to obtain the standard uncertainty of an output
On

41 quantity, such as a molecular weight, requires values and standard uncertainties for the input quantities
42 (such as atomic weights). In the GUM (2), information about input quantities is modelled by means of
43
probability distributions. The Monte Carlo methods of GUM Supplements 1 and 2 (3; 4) use the probability
44
distributions assigned to the input quantities to produce a sample from the probability distribution of the 35
45
ly

46 output quantity. Usually such probability distributions do not depict the variability of the quantity in
47 different materials in nature; rather, these distributions encapsulate the lack of knowledge about the true
48 value of the quantity.
49 In the wording of clause 4.3.7 of the GUM(2), the case of a standard atomic weight can be described
50 as follows. The probability that the value of Ar (E) for some element E lies within the interval from a to b 40
51 is practically 100 %, and the probability that Ar (E) lies outside this interval is essentially zero. If there is
52 no specific knowledge about the possible values of Ar (E) within the interval, one can only assume that it is
53
equally probable for Ar (E) to lie anywhere within it, and therefore be described by a uniform (rectangular)
54
distribution of possible values. This description is well-aligned with the efforts by the CIAAW to establish
55
56 with great care the end points of the atomic-weight interval, or the interval given by a value with the 45
57 uncertainty between parenthesis for the standard atomic weight (17; 9). The wording from the GUM clause
58 4.3.7 makes sufficiently clear under which conditions the rectangular distribution is a legitimate choice and
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 4 of 13

4 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
3 that it applies to standard atomic weights (2). In fact, the use of standard atomic weight implicitly conveys
4 ignorance about the specific provenance or isotopic composition of the material. Because standard atomic
5 weights are “equally applicable to all normal materials”, following Occam’s razor one can assume that it
6 is equally probable for atomic-weight values to lie anywhere within the quoted interval thereby giving no
7 5 preference to any particular material.
8 This recommendation suggests that the standard atomic weight be interpreted probabilistically, taking
9
as viewpoint that the only information at hand that the atomic weight is needed for a normal material.
10
In this conformity, the user of a standard atomic weight is not required to seek a better (and probably
11
12 more accurate) interpretation, taking further information into consideration. It is up to the user of the
13 10 standard atomic weight to decide whether the approach in this recommendation is fit-for-purpose. The
14 accompanying Technical Report (12) describes, and illustrates with examples, assignments of possibly
Fo
15 non-uniform probability distributions to atomic weights that encapsulate additional information about the
16 material (for example, about its provenance). In fact, if the interpretation given in this recommendation is
17 deemed to be not fit-for-purpose, more information should be taken into account to reduce the uncertainty
18 15 about the atomic weight.
rP
19
Hence, under the conditions outlined in this recommendation all standard atomic weights are assigned
20
the rectangular distribution representing a most simplified representation of the lack-of-knowledge (16; 17).
21
22 If Ar (E) is expressed using the interval notation [a, b], then a and b will be the endpoints of the rectangular
distribution. The value of the atomic weight is then given by (2):
ee

23
24 a+b
25 Ar (E) = (1)
2
26
20 which is the mean of the rectangular distribution, and with associated standard uncertainty
rR

27
28 b−a
u(Ar (E)) = √ (2)
29 12
30 which is the standard deviation of the rectangular distribution.
31
ev

In the short-hand notation 95.95(1), it is understood that the value is 95.95 and the number in parenthe-
32
ses is the value of the half-width of the interval referred to the corresponding last digit of the quoted value. In
33 √
34 this example, the half-width of the interval is 0.01 and hence the standard uncertainty is 0.01/ 3 = 0.006.1
ie

35 25 This simplistic interpretation of standard atomic weights – where the midpoint of the standard atomic
36 weight interval is taken as the value, does not reflect that the natural variations of the atomic weight are
37 not necessarily symmetric for the most representative materials (10). The accompanying Technical Report
w

38 (12) provides alternatives to the probabilistic interpretation of these Guidelines to address, among others,
39 the fact that the distribution of values of the atomic weight may not be symmetric. The rectangular
40 30 distribution assigns the same probability density to all values within the standard atomic weight, thus
On

41 giving no preference to any value in particular for the standard atomic weight.
42
43
44
45 5 Examples of obtaining atomic weights
ly

46
47
Performing scientific calculations requires that, at a minimum, all input variables have an associated single
48
49 representative numerical value (for example, the mean or mathematical expectation of the probability
50 35 distribution associated with the input) along with the associated standard uncertainty, and ideally that
51 their associated uncertainties be represented by fully specified probability distributions. In this section,
52 examples are given of how to obtain such representative values from the standard atomic weights.
53
54
55 1 The GUM clause 7.2.2 recommends using this notation for a standard uncertainty with the qualification where the
number in parentheses is the numerical value of (the combined standard uncertainty) uc referred to the corresponding
56
last digits of the quoted result (2). In these guidelines, the conventions of CIAAW are followed. In particular, given the
57
description of the standard atomic weight, as in 192.217(3) for iridium, 0.003 is interpreted as an expanded uncertainty
58 √
with a coverage factor of k = 3.
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 5 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 5

1
2
Table 1. Standard atomic weights of carbon and iridium and their interpretation
3
4
5 Variable Standard atomic weight Assigned Value Standard uncertainty
6 probability
7 distribution
√
8 Ar (C) [12.0096, 12.0116] Rectangular (12.0096+12.0116)/2 = (12.0106 - 12.0096)/(2 3) = 0.0006
9 12.0106
√
10 Ar (Ir) 192.217(3) Rectangular 192.217 0.003/ 3=0.002
11
12
Table 2. Reference atomic weights of oxygen in different sources
13
14
Fo
15 Variable Reference Assigned Value, Standard
16 atomic weight probability Ar (O) uncertainty
17 distribution
√
18 CO in air [15.999 28, 15.999 37] Rectangular (15.999 28+15.999 37)/2 (15.999 37 - 15.999 28)/(2 3) =
rP
19 = 15.999 32 0.000 03
√
20 CO2 in air [15.999 46, 15.999 52] Rectangular (15.999 46+15.999 52)/2 (15.999 52 - 15.999 46)/(2 3) =
21 = 15.999 49 0.000 02
22
ee

23
24 Example 1: Standard atomic weight
25
26
Standard atomic weights are used for normal materials disregarding any specific knowledge about their
rR

27
origin or isotopic composition. To illustrate the use of standard atomic weights, example calculations are
28
29 given for carbon and iridium in Table 1, using equations (1) and (2).
30
31
ev

32 Example 2: Reference atomic weight 5

33
34 There are cases when the uncertainty of a standard atomic weight is not fit for purpose. In such cases, it
ie

35 might be possible to reduce the uncertainty of the atomic weight if the origin of the material is known.
36
For many elements, CIAAW provides detailed information about the atomic weights of elements in various
37
w

38 materials (9; 7). It is important to review the appropriateness of the assigned provenance critically when
39 using reference atomic weights. As an example, Table 2 shows the reference atomic weights of oxygen in 10
40 two sources.
On

41
42
43 Example 3: Atomic weight of boron in specific samples
44
45
ly

When the isotopic composition of the element in the specific material is known, the atomic weight can be
46
calculated directly from this information. Isotopic reference materials fall into this category of materials.
47
48 Often, isotopic composition is measured relative to a standard and is expressed as an isotope delta value 15
49 (9). For an element E with isotopes i E and j E in material P (where usually i > j, meaning that the heavier
50 isotope appears in the numerator of the fraction defining R(i E/j E)P :
51
52 R(i E/j E)P
δ i EP,std = δ(i/j EP,std ) = −1 (3)
53 R(i E/j E)std
54
where R(i E/j E) denotes the isotope amount ratio, n(i E)/n(j E). For boron, most isotope delta measure-
55
56 ments are performed against NIST SRM 951 boric acid standard, which has isotope ratio of boron,
57 R(11 B/10 B)951 = R951 = 4.0436(2)k=2 (7). From here, the isotope ratio of boron in the sample P, 20
58 R(11 B/10 B)P = RP , can be calculated from a measured boron isotope ratio in the test material using
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 6 of 13

6 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
Table 3. Atomic weight of boron in specific samples
3
4
5 Source Isotope delta, Value, Standard uncertainty,
6 δ 11 B951 Ar (B) u(Ar (B))
7 Seawater +39.9(1) ‰ 10.817 88 0.000 07
8 Borax from Turkey –4.1(4) ‰ 10.811 10 0.000 09
9
10
11 equation (3), and then the atomic weight is obtained:
12
13 1 ma (10 B) + ma (11 B)(1 + δ 11 BP,951 )R951
14 Ar (B)P = (4)
mu 1 + (1 + δ 11 BP,951 )R951
Fo
15
16 Equation (4) is obtained by combining equation (3) with the expression below relating the atomic weight
17 to isotope ratios:
18 1 ma (10 B) + ma (11 B)RP
rP
19 Ar (B)P = (5)
mu 1 + RP
20
The uncertainty of Ar (B)P can be propagated from the standard uncertainties of the delta value and that
21
22 5 of the delta-zero standard. Assuming that the uncertainties arising from imperfect knowledge about the
nuclide masses are negligible,
ee

23
24 2
R951 · u2 (δ 11 BP,951 ) + (1 + δ 11 BP,951 )2 · u2 (R951 )
25 u2 (Ar (B)P ) = ∆ma (B) (6)
26 (1 + R951 (1 + δ 11 BP,951 ))4
rR

27
where ∆ma (B) = ma (11 B) − ma (10 B).
28
29
30
31
ev

32
6 Molecular weights
33
34 A common application of atomic weights and standard atomic weights is the calculation of molecular
ie

35 10 weights. These play an important role in relating the SI unit for mass (kg) to the SI unit for amount of
36 substance (mol). The conversion between these two base quantities depends on the chemical substances
37 involved. For a single component in a material or mixture, the molecular weight Mr is conventionally
w

38 calculated from the atomic weights of the elements present in the molecule and the stoichiometric coefficients
39 as
40
X
Mr = vE Ar (E) (7)
On

41 E
42
43 15 where vE denotes the stoichiometric coefficient and Ar (E) is the atomic weight of element E. The summation
44 runs over all elements E present in the molecule. For example, ethanol (C2 H6 O) is composed of carbon,
45 hydrogen and oxygen with stoichiometric coefficients 2, 6, and 1, respectively.
ly

46 Just as in the case of atomic weights, the measurement uncertainty associated with the molecular weight
47 of a molecule can be evaluated using the law of propagation of uncertainty from the GUM (2), or the Monte
48 20 Carlo method from the GUM-S1 (3). Usually the uncertainty calculation for the molecular weight will be
49 part of a larger calculation, such as the computation of the mixture composition from a gravimetrically
50
prepared mixture (15; 8). Whether the methods for calculating the measurement uncertainty associated
51
with a molecular weight of these Guidelines are used, or those of the accompanying Technical Report (12),
52
53 should be decided on the impact that the calculated uncertainties have in the subsequent calculations.
54 25 Applying the law of propagation of uncertainty to the expression for calculating the uncertainty of the
55 molecular weight X
56 u2 (Mr ) = 2 2
vE u (Ar (E)) (8)
57 E
58 where the correlation between pairs of atomic weights of elements can be neglected.
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 7 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 7

1
2
Table 4. Uncertainty budget for the standard molecular weight of ethanol
3
4
5 Variable Standard Assigned Value Standard Stoichiometric Uncertainty
6 atomic weight probability uncertainty coefficient contribution
7 distribution
8 Ar (C) [12.0096, 12.0116] Rectangular 12.0106 0.000 58 2 0.001 15
9 Ar (H) [1.007 84, 1.008 11] Rectangular 1.007 975 0.000 078 6 0.000 47
10 Ar (O) [15.999 03, 15.999 73] Rectangular 15.9994 0.000 21 1 0.000 21
11 Mr (C2 H6 O) 46.0685 0.001 30
12
13
14
7 Examples of obtaining molecular weights
Fo
15
16
17 In this section, illustrative examples are given for the computation of standard and reference molecular
18 weights of ethanol.
rP
19
20
21
22
Example 4: Standard molecular weight of ethanol
ee

23
24 To illustrate the use of the equations for calculating the molecular weight, the molecular weight of ethanol 5
25 (C2 H5 OH or C2 H6 O) is calculated as shown in Table 4.
26 The expected values of the standard atomic weights are taken as the midpoints of the corresponding
rR

27 standard-atomic-weight intervals. As for the uncertainty, the width of these intervals is converted into
28 standard uncertainties using expressions applicable to rectangular distributions over such intervals. The
29 uncertainty contribution is the product of the stoichiometric coefficient and the standard uncertainty. These 10
30
are squared, summed, and then the square root is taken to compute the standard uncertainty associated
31
ev

with the molecular weight (2).

32
33 From the magnitudes of the uncertainty contributions (Table 4, last column), it can be seen that the
34 largest contributor to the standard uncertainty of the molecular weight is carbon whereas the contribution
ie

35 of oxygen is approximately ten times smaller. The presentation of the uncertainty budget in this format 15
36 facilitates the comparison of the relative importance of the different sources of uncertainty. In this case it
37 indicates clearly that if the uncertainty associated with the molecular weight of ethanol is deemed to be
w

38 too large to be fit-for-purpose, then one has to find a more precise value for the atomic weight of carbon.
39 This can be achieved, for example, if the source of the ethanol is known, and it is illustrated in the next
40
example. 20
On

41
42
43
44 Example 5: Reference molecular weight of ethanol
45
ly

46 There are situations in science and technology where the uncertainties associated with the standard atomic
47 weights are prohibitively large. In such situations, it might be possible to reduce the uncertainty associated
48 with the atomic weights by considering the origin of the material at-hand. For example, the atomic weight of
49 the oxygen in air or seawater is known with smaller uncertainty than the standard atomic weight of oxygen 25
50
applicable to all natural materials. To illustrate the use of CIAAW diagrams detailing the natural variations
51
52 of the isotopic composition of the elements (9), consider naturally-produced ethanol whose hydrogen and
53 oxygen comes from seawater and air, and whose carbon derives from corn. The following reference atomic
54 weights of carbon, hydrogen, and oxygen were obtained:
55 Ar (C) = [12.0107, 12.0111] for naturally-occurring ethanol, 30
56 Ar (H) = [1.007 90, 1.008 01] for naturally-occurring water,
57 Ar (O) = [15.999 05, 15.999 45] for continental water and air.
58
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 8 of 13

8 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
Table 5. Uncertainty budget for the reference molecular weight of natural ethanol
3
4
5 Variable Expected Assigned Standard Stoichiometric Uncertainty
6 value probability distribution uncertainty coefficient contribution
7 Ar (C) 12.0109 Rectangular 0.000 12 2 0.000 24
8 Ar (H) 1.007 955 Rectangular 0.000 032 6 0.000 19
9 Ar (O) 15.999 25 Rectangular 0.000 12 1 0.000 12
10
11
12 For these atomic weights, the following expected values are derived:
13
Ar (C) = (12.0111 − 12.0107)/2 = 12.0109,
14
Ar (H) = (1.008 01 − 1.007 90)/2 = 1.007 955,
Fo
15
16 Ar (O) = (15.999 45 − 15.999 05)/2 = 15.999 25.
17
18 5 Table 5 shows the reference molecular weight of ethanol calculated from these reference atomic weights. One
rP
19 can see that the reference molecular weight of ethanol has an uncertainty that is four times smaller than
20 that of the standard molecular weight (Table 4). If the uncertainty associated with the reference molecular
21 weight is still deemed to be too large to be fit-for-purpose, then the atomic-weights of all elements present
22 should be measured.
ee

23
24
25
26 10 8 Monte Carlo methods
rR

27
28
In addition to the law of propagation of uncertainty from the GUM (2), the Monte Carlo method of the
29
30 GUM-S1 may also be used to propagate the uncertainty associated with the standard atomic weights in
31 the calculation of molecular weights. It can be described by the following steps (3):
ev

32
33 1. Assign a rectangular probability distribution to each of the standard atomic weights to be used for the
34 15 calculation of the molecular weight of the molecule of interest;
ie

35 2. Sample an atomic-weight value for each element from above distributions;

36 3. Compute the molecular weight using the values obtained in step 2;
37 4. Repeat steps 2 and 3 as many times as necessary to get a sufficiently large number of values for the
w

38 molecular weight.
39
40
On

41
20 These molecular-weight values are a sample from the probability distribution of the molecular weight
42
43 and they can be used to compute the value, standard uncertainty, and coverage interval, or they may
44 be used directly in subsequent Monte Carlo sampling. To illustrate the process, the standard molecular
45 weight is calculated for carbon dioxide (CO2 ), ethanol (C2 H6 O), ethanethiol (C2 H6 S), and methylamine
ly

46 (methanamine, CH3 NH2 ).

47 25 Figure 1 shows the results obtained by application of the Monte Carlo method of the GUM-S1 when
48 rectangular distributions are assigned to the atomic weight intervals, and in particular reveals that only for
49 methylamine is the distribution of the molecular weight approximately normal (or, Gaussian) The Monte
50
Carlo results are summarized in Table 6.
51
The differences between the GUM-S1 coverage intervals and those based on the mean and standard
52
53 30 uncertainty computed from the law of propagation of uncertainty become more pronounced as the coverage
54 probability increases.
55 For homoatomic molecules such as O2 , P4 , S8 or C60 , the probability distribution of the standard
56 molecular weight is rectangular which results from rescaling the corresponding rectangular distributions of
57 the standard atomic weights.
58
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 9 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 9

1
2
3
4
5
6
7
8
9
10
11
12
13
14
Fo
15
16
17
18
rP
19
20
21
22
ee

23
24
25
26
rR

27
28
29
Fig. 1. Propagation by Monte Carlo simulation of the probability distributions of the standard atomic weights to calculate
30
the probability distribution of the molecular weight of carbon dioxide, ethanol, ethanethiol, and methylamine. One million
31
ev

samples were used to generate the output probability distributions.

32
33
34
ie

35
36
37
w

38
39
40
On

41
42
43
44
Table 6. Standard molecular weights, their standard uncertainties, lower- and upper-values of the 95 % coverage intervals,
45
ly

and the relative widths of these intervals for four molecules

46
47
48 Molecule Mr u(Mr ) q0.025 (Mr ) q0.925 (Mr ) ∆q/u(Mr )
49 CO2 44.0094 0.0007 44.0081 44.0108 3.77
50 C2 H5 OH 46.0685 0.0013 46.0662 46.0707 3.61
51 C2 H5 SH 62.1366 0.0051 62.1278 62.1453 3.44
52 CH3 NH2 31.0573 0.0007 31.0559 31.0587 3.79
53
54
55
56
57
58
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 10 of 13

10 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
3 9 Combining molecular weights for components in mixtures
4
5
In chemistry, not one but several molecular weights are often required simultaneously. For example, gravi-
6
7 metric determination of the mass fraction of silver by the precipitation of silver chloride requires the ratio
8 Ar (Ag)/Mr (AgCl). In mixtures like air or natural gas, often there are components that have certain ele-
9 5 ments in common. The fact that such elements appear in multiple molecules gives rise to correlations in the
10 computed molecular weights (12; 15). In applications where the uncertainty associated with the molecular
11 weight matters, such correlations may matter too.
12
13
14
Example 6: Ratio of the standard atomic weights of oxygen and hydrogen
Fo
15
16
17 Atomic weights often are used in the form of ratios. An important historic example of this was the ratio of
18 10 the standard atomic weights of oxygen and hydrogen, once thought to be exactly sixteen (14):
rP
19
Ar (O)
20 R= (9)
21 Ar (H)
22 Since the standard atomic weights of hydrogen and oxygen can be assumed to be uncorrelated, the following
ee

23 equation can be used:

24 u2 (R) u2 (Ar (O)) u2 (Ar (H))
25 ≈ + (10)
R2 A2r (O) A2r (H)
26
rR

27 The equality is only approximate because, differently from what happens for molecular weights, which are
28 linear combinations of atomic weights, the evaluation produced by the law of propagation of uncertainty
29 15 provides only an approximation to the uncertainty associated with a ratio, as well as for other non-linear
30 measurement functions [9, clause 5.1.6].
31
ev

The correlation between the atomic weights is a complex issue. On a deeper level, for example, it is
32 known that the natural variations of the atomic weight of hydrogen and oxygen track each other. However,
33
if such details become relevant, one would not use standard atomic weights in the calculations, but rather
34
20 take the available data and construct a more sophisticated measurement model (12).
ie

35
36 Using the 2013 standard atomic weights R = 15.8728 with u(R) = 0.0012 which is significantly
37 different from 16. In this example we considered two independent atomic weights. Next we will consider a
w

38 more complex situation in which the two components share a common element.
39
40
On

41 Example 7: Composition of a mixture of methane and propane

42
43
25 The molecular weight of simple molecules is rarely the ultimate quantity of interest; more often than not it
44
45 is a part of a much larger calculation, aiming, for example, at the calculation of a mixture composition. In
ly

46 this vein, let us consider a mixture of pure methane and pure propane, which can be viewed as a simplified
47 model of a synthetic natural gas (15).
48 The uncertainties of the standard molecular weights of methane and propane are calculated as before.
49 30 For methane, the variance associated with the standard molecular weight is
50
51 u2 (Mr,CH4 ) = 12 u2 (Ar (C)) + 42 u2 (Ar (H))
52 = 1 · (0.00058)2 + 16 · (0.000078)2 = 4.3 · 10−7 (11)
53
54 and the standard uncertainty is u(MCH4 ) = 0.000 66. For propane, the variance associated with the
55 molecular weight is
56
57 u2 (Mr,C3 H8 ) = 32 u2 (Ar (C)) + 82 u2 (Ar (H))
58
= 9 · (0.00058)2 + 64 · (0.000078)2 = 3.4 · 10−6 (12)
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 11 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 11

1
2
3 and the standard uncertainty is u(Mr,C3 H8 ) = 0.001 84.
4 The covariance between the molecular weights of methane (CH4 ) and propane (C3 H8 ) can be computed
5 using equation F.2 of the GUM (2):
6 X
7 u(Mr,CH4 , Mr,C3 H8 ) = vz,CH4 · vz,C3 H8 · u2 (Ar (z)) (13)
8 z=C,H

9 where the summation runs over all elements in common (in this example z = 2, namely carbon and
10 hydrogen) and vz,X is the stoichiometric coefficient of element z in molecule X. For the molecular weights 5
11
of methane and carbon, this implies that the summation covers carbon and hydrogen. The expression for
12
13 the covariance between the molecular weights becomes
14 u(Mr,CH4 , Mr,C3 H8 ) = 1 · 3u2 (Ar (C)) + 4 · 8u2 (Ar (H))
Fo
15
16 = 3 · (0.00058)2 + 32 · (0.000078)2 = 1.19 · 10−6 (14)
17 The covariance between the molecular weights of methane and propane may seem a small number, but the
18 correlation coefficient, r(Mr,CH4 , Mr,C3 H8 ), reveals strong correlation between the two molecular weights:
rP
19
20 u(Mr,CH4 , Mr,C3 H8 ) 1.19 · 10−6
21 r(Mr,CH4 , Mr,C3 H8 ) = = = 0.989 (15)
u(Mr,CH4 )u(Mr,C3 H8 ) 0.00066 · 0.00184
22
The calculation of a correlation coefficient from a covariance is given in the equation 14 of the GUM (2). 10
ee

23
24 To appreciate the significance of the correlation between the standard molecular weights of methane
25 and propane, consider a mixture of these two gases which is prepared from pure methane (mCH4 = 459.6263
26 g) and pure propane (mC3 H8 = 140.3737 g) (15; 8). The amount fraction of propane in the mixture is given
rR

27 by
28 nC3 H8 mC3 H8 /Mr,C3 H8
29 xC3 H8 = = (16)
nCH4 + nC3 H8 mCH4 /Mr,CH4 + mC3 H8 /Mr,C3 H8
30
31 If we consider the masses of the parent gases to be free from uncertainty, just as their compositions, the 15
ev

32 squared standard uncertainty of xC3 H8 can be expressed as

33
∂xC3 H8 2 2 ∂xC3 H8 2 2
   
34 u2 (xC3 H8 ) = u (Mr,CH4 ) + u (Mr,C3 H8 )
ie

35 ∂Mr,CH4 ∂Mr,C3 H8
36 
∂xC3 H8

∂xC3 H8

37 +2 u(Mr,CH4 , Mr,C3 H8 ) (17)
w

∂Mr,CH4 ∂Mr,C3 H8
38
39 The expressions for the sensitivity coefficients (partial derivatives) are as follows (15):
40
∂xC3 H8 mC3 H8 /Mr,C3 H8 mCH
On

41 = · 2 4
∂Mr,CH4 2
(mCH4 /Mr,CH4 + mC3 H8 /Mr,C3 H8 ) Mr,CH
42 4
43 140.3737/44.0956 459.6263
44 = · = 0.005610 (18)
(459.6263/16.0425 + 140.3737/44.0956)2 16.04252
45
ly

46 ∂xC3 H8 mCH4 /Mr,CH4 mC H

=− · 23 8
47 ∂Mr,C3 H8 (mCH4 /Mr,CH4 + mC3 H8 /Mr,C3 H8 )2 Mr,C H 3 8
48
459.6263/16.0425 140.3737
49 =− · = −0.002041 (19)
(459.6263/16.0425 + 140.3737/44.0956)2 44.09562
50
51 The amount fraction propane in the mixture is xC3 H8 = 0.100 00 and the associated variance is:
52
53 u2 (xC3 H8 ) = 0.0056102 · 0.000662 + (−0.002041)2 · 0.001842
54 + 2 · 0.005610 · (−0.002041) · 1.19 · 10−6 = (5.6 · 10−7 )2 (20)
55
56 By comparison, neglecting the covariance between the standard molecular weights of CH4 and C3 H8 yields 20
57 the following variance of xC3 H8 :
58
u2 (xC3 H8 ) = 0.0056102 · 0.000662 + (−0.002041)2 · 0.001842 = (5.3 · 10−6 )2 (21)
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
 IUPAC Page 12 of 13

12 A.M.H. van der Veen, Uncertainty of atomic weights

1
2
Table 7. Correlation coefficients between the standard molecular weights of common substances
3
4
5 Molecule H2 O CO2 NO2 CH4 SO2
6 H2 O 1 0.481 0.700 0.280 0.070
7 CO2 1 0.516 0.707 0.052
8 NO2 1 0 0.075
9 CH4 1 0
10 SO2 1
11
12
13 The difference in the computed standard uncertainties is significant; ignoring the covariance between the
14 molecular weights results in a ten-fold greater uncertainty.
Fo
15
The Monte Carlo method of the GUM-S1 can take correlations into account to reproduce the results
16
17 above with much greater ease, for example in these few lines of R code (13):
18 5 H = runif (1 e6 , min =1.00784 , max =1.00811)
rP
19 C = runif (1 e6 , min =12.0096 , max =12.0116)
20 M . methane = C + 4 * H ; M . propane = 3 * C + 8 * H
21 m . propane = 140.3737; m . methane = 459.6263
22 x . propane = ( m . propane / M . propane ) / (( m . methane / M . methane ) + ( m . propane / M . propane ) )
10
ee

23 c ( x . propane = signif ( mean ( x . propane ) ,5) , " u ( x . propane ) " = signif ( sd ( x . propane ) , 2) )
24
25 The corresponding output is
26
rR

27 x . propane u ( x . propane )
28 1.0 e -01 5.6 e -07
29
30 For mixtures of multiple components, this pair-wise evaluation of the correlations is too cumbersome.
31
ev

15 Instead, the model for calculating the set of molecular weights can be viewed as a multivariate measurement
32
model, and the measurement uncertainty can be evaluated using the methods of GUM-S2 (4). An example
33
34 for a hydrocarbon mixture is given elsewhere (12). It has been demonstrated how to use this approach
ie

35 in computing the composition of a mixture on a molar basis using the law of propagation of uncertainty
36 (15) and the Monte Carlo method. Table ?? shows correlation coefficients between the standard molecular
37 20 weights of five common substances.
w

38
39
40
Membership of sponsoring bodies
On

41
42
43 Membership of the IUPAC Inorganic Chemistry Division Committee for the period 2014–2015 was as
44 follows:
45
ly

President: J. Reedijk (Netherlands); Secretary: M. Leskelä (Finland); Vice President: L. R. Öhrström

46
25 (Sweden); Past President: R. D. Loss (Australia); Titular Members: T. Ding (China); M. Drábik
47
48 (Slovakia); D. Rabinovich (USA); E. Tshuva (Israel); T. Walczyk (Singapore); M. Wieser (Canada); As-
49 sociate Members: J. Buchweishaija (Tanzania); J. García-Martínez (Spain); P. Karen (Norway); A.
50 Kiliç (Turkey); K. Sakai (Japan); R.N. Vannier (France); National Representatives: Y. Abdul Aziz
51 (Malaysia); L. Armelao (Italy); A. Badshah (Pakistan); V. Chandrasekhar (India); J. Galamba Correia
52 30 (Portugal); S. Kalmykov (Russia); S. Mathur (Germany); L. Meesuk (Thailand); B. Prugovec̆ki (Croatia);
53 N. Trendafilova (Bulgaria)
54
55 Membership of the Commission on Isotopic Abundances and Atomic Weights for the period 2014–2015 was
56 as follows:
57 Chair: J. Meija (Canada); Secretary: T. Prohaska (Austria); Titular Members: W.A. Brand (Ger-
58
35 many); M. Gröning (Austria); R. Schönberg (Germany); X. K. Zhu (China); Associate Members: T. Hi-
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700
Page 13 of 13 IUPAC

A.M.H. van der Veen, Uncertainty of atomic weights 13

1
2
3 rata (Japan); J. Irrgeher (Austria); J. Vogl (Germany); National Representatives: T.B. Coplen (USA);
4 P. De Bièvre (Belgium); Ex-officio: J. Reedijk (Netherlands).
5
6 Acknowledgment: The following IUPAC projects contributed to this Recommendation: 2011-040-2-200
7 and 2013-032-1-200.
8
9
10
11 References 5
12
13 [1] Bièvre, P. D. and H. Peiser (1992): “‘atomic weight’ — the name, its history, definition, and units,” Pure Appl Chem,
14 64, 1535–1543.
Fo
15 [2] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, and OIML (2008): Guide to the Expression of Uncertainty in Measure-
16 ment, JCGM 100:2008, GUM 1995 with minor corrections, BIPM.
17 [3] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, and OIML (2008): Supplement 1 to the ‘Guide to the Expression of 10
18 Uncertainty in Measurement’ – Propagation of distributions using a Monte Carlo method, JCGM 101:2008, BIPM.
rP
19 [4] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, and OIML (2011): Supplement 2 to the ‘Guide to the Expression of
20 Uncertainty in Measurement’ – Extension to any number of output quantities, JCGM 102:2011, BIPM.
21 [5] Brand, W. A. (2012): “Atomic weights: not so constant after all,” Anal Bioanal Chem, 405, 2755–2761.
22 [6] Cohen, E., T. Cvitas, J. Frey, B. Holmström, K. Kuchitsu, R. marquardt, I. Mills, F. Pavese, M. Quack, J. Stohner, 15
H. Strauss, M. Takami, and A. Thor (2008): Quantities, Units and Symbols in Physical Chemistry: RSC, Cambridge:
ee

23
IUPAC and Royal Society of Chemistry, 3 edition.
24
[7] Coplen, T., J. Böhlke, P. D. Bièvre, T. Ding, N. Holden, J. Hopple, H. Krouse, A. Lamberty, H. Peiser, K. Révész,
25
S. Rieder, K. Rosman, E. Roth, P. Taylor, R. V. Jr., and Y. Xiao (2002): “Isotope-abundance variations of selected
26 elements,” Pure Appl Chem, 74, 1987–2017. 20
rR

27 [8] ISO (2001): ISO 6142 Gas analysis - Preparation of calibration gas mixtures - Gravimetric methods, Geneva: ISO.
28 [9] Meija, J., T. B. Coplen, M. Berglund, W. A. Brand, P. D. Bièvre, M. Gröning, N. E. Holden, J. Irrgeher, R. D. Loss,
29 T. Walczyk, and T. Prohaska (2016): “Atomic weights of the elements 2013,” Pure Appl Chem, 88.
30 [10] Meija, J. and Z. Mester (2011): “Comment on the uncertainties in isotope patterns of molecules,” Analytica Chimica
31 Acta, 694, 174–176. 25
ev

32 [11] Peiser, H., N. Holden, P. D. Bièvre, I. Barnes, R. Hagemann, J. de Laeter, T. Murphy, E. Roth, M. Shima, and
33 H. Thode (1984): “Element by element review of their atomic weights,” Pure Appl Chem, 56, 695–768.
34 [12] Possolo, A., A. van der Veen, J. Meija, and B. Hibbert (2016): “On the interpretation of the uncertainty of the
ie

35 atomic weights,” Pure Appl Chem, in preparation.

36 [13] R Core Team (2015): R: A Language and Environment for Statistical Computing, R Foundation for Statistical Com- 30
37 puting, Vienna, Austria, URL https://www.R-project.org/.
w

38 [14] Richards, T. (1898): “A table for atomic weights,” Proc Am Acad Arts and Sci, 33, 293–302.
[15] van der Veen, A. M. H. and K. Hafner (2014): “Atomic weights in gas analysis,” Metrologia, 51, 80–86, URL http:
39
//dx.doi.org/10.1088/0026-1394/51/1/80.
40
[16] Wieser, M. E. and M. Berglund (2009): “Atomic weights of the elements 2007,” Pure Appl Chem, 81. 35
On

41
[17] Wieser, M. E. and T. B. Coplen (2011): “Atomic weights of the elements 2009,” Pure Appl Chem, 83, 359–396.
42
43
44
45
ly

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
P.O. 13757, Research Triangle Park, NC (919) 485-8700