Beruflich Dokumente
Kultur Dokumente
A few level approach for the electronic partition function of atomic systems
Gianpiero Colonna a,⁎, Mario Capitelli b
a
CNR IMIP Bari Via Amendola 122/d, 70126, Bari, Italy
b
Department of Chemistry, University of Bari, Via Orabona 4, 70126, Bari, Italy
a r t i c l e i n f o a b s t r a c t
Article history: A simplified model to calculate partition functions and thermodynamic properties of atomic species is
Received 18 March 2009 presented. This model consists in grouping the atomic states in few virtual levels. Their statistical weights
Accepted 1 July 2009 and energies are calculated summing or mediating over the states belonging to each group. The partition
Available online 16 July 2009
function is calculated considering the virtual levels to follow the Boltzmann distribution. A theoretical
foundation of the model has been described for a general case and verified for hydrogen, oxygen and nitrogen
PACS:
atoms. Two- and three-level models are adequate to keep the error within a few percent. A second order
51.30. + i
52.38.Mf
correction term can also be added to further reduce the error without changing the advantages of the model.
© 2009 Elsevier B.V. All rights reserved.
Keywords:
Atomic partition function
Statistical thermodynamic
Superconfiguration
0584-8547/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.sab.2009.07.002
864 G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873
εsl b εM εsl
s
− where 〈·〉 is the classical weighted statistical mean. Under these
Qs = ∑ gsl e T ð11Þ
l=0 assumption the mean energy (see Eqs. (4)–(6)) is given by
where ε
⋆ 3 1 − ε 3 1 −z
U = T + ⋆ Gεe T = + ⋆ Gze ð20Þ
2 Q z 2 Q
M
εs = Is −Δs ð12Þ
and the specific heat (see Eqs. (7) and (8)) by
Is being the ionization potential and Δs the energy cutoff.
Some databases reporting partition functions and other internal 2 ε ε 2 ε 2
⋆ 3 Gε − ⋆ − 3 Gg ε − 3 Gg0 z −z
thermodynamic properties for different cutoffs are available [14,24]. It Cv = + 2 ⋆2 e T Q −Ge T = + 2 0 ⋆2 e T = + ⋆2
e
2 T Q 2 T Q 2 Q
must be pointed out that in real systems the cutoff is not constant but
it depends on some quantities such as particle density (Fermi ð21Þ
criterion) or electron density (Griem criterion). As a consequence,
the cutoff can be determined once the plasma composition is known. The demonstration of Eqs. (17)–(21) starts rewriting Eq. (17) by
This approach increases the computational time necessary to separating the contribution of the ground states from excited levels
ε
−
determine the atomic properties, especially when kinetic or thermo- and by multiplying and dividing the last term by the factor e T
approximating the internal partition function as being τl the truncation error, and calculating the sum over the levels
Nsgr εsk
−
ε
εl −ε −
ε
G −
ε
−
ε
⋆ −
Qs ≈Qs = ∑ Gsk e T ð14Þ Q = g0 + Ge T 〈1− + τl 〉 = g0 + Ge T − e T ð〈ε〉−εÞ + Ge T 〈τ〉:
T T
k=0
ð24Þ
where the number of groups (Ngr s ) is much smaller than the number of
levels. As we will show later, the group statistical weight (Gsk) and Being the ε = 〈ε〉 by definition (see Eq. (19)), the third term in Eq. (24)
energy (Esk), given by is null, obtaining
ε ε ε
Lfin − − ⋆ −
Gsk = ∑ gsl
sk
ð15Þ Q = g0 + Ge T + Ge T 〈τ〉 = Q + Ge T 〈τ〉 ð25Þ
l = Lin
sk
are good approximations under the condition that the mean energy of This result shows that Eq. (17) is accurate at the first order. Some
the group is much higher than the level energy variance. The symbols considerations can explain why this approximation works. Supposing
fin
Lin
sk and Lsk are the first and last level belonging to the k-th group. This that the level dispersion around the mean energy is small with respect
method is very useful because the quantities E and G do not depend to the mean energy. The Taylor expansion can be used only if the
on the temperature and only the influence of the cutoff must be taken temperature is not too low. In particular this approximation can be
into account. safely used if the temperature is of the same order as the mean energy
Let's consider the simple case of two groups of levels, the first one or higher. On the other hand, for temperature much lower than
includes only the ground state, with zero energy, and the second one all the mean energy, the exponential is very small, and therefore also
the excited levels. To simplify the notation the group and species indexes in this case the expression is accurate, even if the Taylor series
are omitted for the rest of this section. We want to demonstrate that truncated at the first term cannot be used. The value of the error can
be limited by the following relations. For the properties of the ex-
εl ε
N
− ⋆ − −z ponential function
Q = ∑ gl e T ≈Q = g0 + Ge T = g0 + Ge ð17Þ
l=0
δQ N 0 ð27Þ
ε
where z = and ε and G are calculated following the Eqs. (15)
T
and (16) i.e.
and, considering the levels ordered by increasing energies
N
εN ε1
G = ∑ gl ð18Þ − ⋆ −
l=1 g0 + Ge T ≤Q ≤Q≤g0 + Ge T ð28Þ
866 G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873
1 2
τl ≈ ðεl −εÞ ð30Þ
2T 2
ε ε
G −T 2 G −T 2 G 2 −z 2 ⋆
δQ ≈ e 〈ðεl −εÞ 〉 = e σε = z e δε = δQ ð31Þ
2T 2 2T 2 2
σε
δε = ð32Þ
ε
This means that to the very low and very high temperature, as already
discussed, the error tends to be null, and the maximum error is δU ⋆ dQ dQ ⋆ 1 dQ dQ ⋆ dδQ dδQ ⋆
observed for temperature values of the same order of the mean =− − ⋆ ≈− ⋆ − =− ⋆ ≈ ⋆ ð37Þ
ε Qdz Q dz Q dz dz Q dz Q dz
energy. The maximum error of the partition function is of the order of ⋆
δQ z−2
the square of the relative level energy dispersion and the total = ⋆
Q z
statistical weight.
The relative error can be approximated as
G
which is almost proportional to δ2ε , where α = . The plot of the
g0
relative error of partition function, normalized by the square of the
dispersion has been reported in Fig. 1. We can observe that for high
temperature, the relative error is independent of α showing a power
dependence on z. Close to the maximum it depends weakly on α,
decreasing rapidly.
The maximum of the relative error is calculated finding the roots of
the derivative
⋆
d δQ ⋆ 1 dδQ ⋆ ⋆ dQ 1 ⋆ dδQ
⋆
⋆ dQ
= 2 Q −δQ = 2 Q −δQ
dz Q Q dz dz Q dz dz
⋆
1 −z δQ ⋆ −z ð36Þ
= 2 ðg0 + Ge Þ ð2−zÞ + δQ Ge
Q z
⋆ ⋆
δQ −z g0 δQ −z
= ½g ð2−zÞ + 2Ge = ½ð2−zÞ + 2αe = 0:
zQ 2 0 zQ 2
ε
The root of Eq. (36) zM = ⋆ can be found numerically and it
T
depends only on α. In Fig. 2 we have reported the root of Eq. (36),
which gives the temperature of maximum error, and the correspond-
ing value of the relative error, reduced by the factor δ2ε as a function of Fig. 2. Value of zM = Tε⋆ (root of Eq. (36)) and the corresponding maximum value of the
the ratio α. It can be observed that these quantities depends on log α partition function relative error (reduced by the factor δ2ε ) as a function of the ratio
and that their values are not very large. In fact if the energy dispersion α = gG0 .
G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873 867
d δU ⋆ −z 2 2
= 0⇒αe ð4z + 5z−12Þ−3ðz −5z + 4Þ = 0 ð39Þ
dz U⋆
and the values of the maximum error are reported in Fig. 4. It must be
noted that the maximum of the internal energy relative error is of the
same order as the maximum of the relative error of the partition
function (see Fig. 2).
For the specific heat, due to the additive property, it is
2 ⋆ ⋆
⋆ int⋆ z dδU δQ ⋆
δCv = δCv =− = ⋆2 ½2ð1−zÞQ + g0 zðz−2Þ ð40Þ Fig. 4. Value of zM =
ε
(root of Eq. (39)) and the corresponding internal energy relative
ε dz Q T⋆
G
error reduced by the factor δ2ε as a function of the ratio α = .
g0
δCv⋆ ⋆
⋆ 2ð1−zÞQ + g0 zðz−2Þ in Fig. 6. Compared with the relative error of the partition function
= 2δQ : ð41Þ
Cv⋆ 3Q + 2g0 Gz2 e−z
⋆2 (see Fig. 2) and of the internal energy (see Fig. 4) the one of the
specific heat is 4 times larger.
We have shown that the first order approximation results in a
This expression, divided by the level energy dispersion, has been
simple reduced level representation. It must be stressed that
plotted as a function of z in Fig. 5. It can be observed the function trend
increasing the order of approximation does not increase the
is more complex than the previous quantities, as well as the analytical
complexity of the simplified scheme, and, furthermore, uses
expression of its derivative. Moreover two different maxima are
coefficients as σε that do not depend on the temperature, keeping
observed, the second is larger than the first one at low α but this
the same advantages as the first order approximation, but reducing
relation is inverted as α increases. Both maxima have been calculated
evaluating the derivative numerically. The results have been reported
Fig. 5. Value of the relative error of the constant volume specific heat (see Eq. (41))
Fig. 3. Value of the relative error of the internal energy U (see Eq. (38)) reduced by the reduced by the factor δ2ε as a function of z for different values of α. The value of α is
factor δ2ε as a function of z for different values of α. The value of α is varied by a factor 10. varied by a factor 10.
868 G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873
Fig. 6. Position zM of the two maxima (see Fig. 5) (lower frame) and the corresponding
relative error reduced by the factor δ2ε (upper frame) for constant volume specific heat
G
as a function of the ratio α = .
g0
8 ⋆⋆ ⋆ ⋆
< Q = Q + δQ
>
⋆⋆ ⋆ ⋆
U = U + δU : ð42Þ Fig. 7. Total statistical weight (G), mean energy (E [eV]) and square of the relative level
>
: ⋆⋆ energy dispersion (δ2ε ) (see Eqs. (45) and (47)) of excited levels as a function of the
⋆ ⋆
Cv = Cv + δCv maximum value of the principal quantum number (N) for hydrogen atoms.
Table 1
Statistical weight, energy (eV) and electronic configuration of the first levels of H, N and O atoms.
Fig. 8. Relative error of the H partition function vs. temperature for different values of
the maximum value of the principal quantum number. Legend: ⁎ - calculated with
Eq. (35); 1 - exact error with the first order approximation; 2 - exact error with the
second order approximation.
Fig. 9. Relative error of the H internal energy vs. temperature for different values of the
Comparing this result with that reported in Fig. 2, we can assert that maximum value of the principal quantum number. Legend: ⁎ - calculated with Eq. (35);
the relative error of the partition function, decreasing with the 1 - exact error with the first order approximation; 2 - exact error with the second order
number of levels if N is large enough, is limited to 2%. approximation.
870 G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873
Table 2
Number of levels and group data as a function of the cutoff for H (Ng1st = 1), N (Ng1st = 3)
g
and O (N1st = 5) atoms.
Fig. 11. Maximum relative error of the H thermodynamic quantities as a function of the Fig. 12. Absolute value of the first order relative error of nitrogen thermodynamic
number of excited levels. Legend: ⁎ - calculated with Eq. (35); 1 - exact error with the functions versus temperature, calculated considering Δs = 500 cm− 1. The different
g
first order approximation; 2 - exact error with the second order approximation. curves refers to the values of δ1st.
G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873 871
Table 3
Maximum of 1st and 2nd order percentage error, as a function of the initial group level, for N and O thermodynamic quantities.
g
N1st Nitrogen Oxygen
1st order 2nd order 1st order 2nd order
Q U [eV] Cv Q U [eV] Cv Q U [eV] Cv Q U [eV] Cv
1 46 29 70 41 28 70 52 32 62 50 33 73
3 2.0 1.3 2.8 0.52 0.39 0.91 18 10 19 15 9.7 20
5 1.3 0.85 1.8 0.28 0.21 0.49 1.7 1.3 2.7 0.41 0.37 0.86
7 0.99 0.64 1.4 0.18 0.14 0.31 1.1 0.80 1.7 0.18 0.16 0.38
9 0.64 0.42 0.89 0.094 0.072 0.16 0.87 0.65 1.4 0.13 0.12 0.28
The maximum has been calculated with respect to temperature and number of levels.
g
function. As an example the lumped level which includes all nitrogen c) N1st = 10
levels from i = 3 to N = 413, the maximum level of our database
2:38396 3:57561 11:9955 ε10
compatible with a lowering of the ionization potential of 250 cm− 1, − − − −
Q = 4 + 10e T + 6e T + … + 4e T + G10 e T
has a total statistical weight G = 51594 and average energy
E = 14.4602 eV. Note that the G values dramatically decrease with ð51Þ
increasing Δs. In the same table we have also reported the corre- where Gi and εi are calculated starting the sum in Eqs. (18) and (19)
sponding values for atomic hydrogen; in this case N corresponds to from l = i
the principal quantum number of the last considered level.
Fig. 12 reports the first order relative error on the partition N
function, internal energy and specific heat of nitrogen system with a Gi = ∑ gl
l=i
parametric cutoff of 500 cm− 1, which approximately corresponds to : ð52Þ
an LTE (Local Thermodynamic Equilibrium) atmospheric nitrogen 1 N
εi = ∑ gε
plasma. The three curves, reported in the figure, correspond to the Gi l = i l l
level from which arises the lumped level. So
g
g we can expect that the calculation with N1st = 1 will give unsatisfactory
a) N1st = 1 means that we are calculating the partition function (and 2
results due to the insertion of the P states in the lumped level, while
related thermodynamic properties) of atomic nitrogen as g
N1st = 3 case should give satisfactory results as in the case of atomic
g
ε1 hydrogen. We also expect an increase of accuracy with N1st = 10. This is
−
Q = 4 + G1 e T ð49Þ indeed the case as can be understood by inspection of Table 3. We can
g
see that the N1st = 1 approximation gives not acceptable errors for all
g
g
b) N1st = 3 quantities in all the temperature range, while N1st = 3 approximation
g
gives errors below 1%. The case with N1st = 10 refines the approximation
2:38396 3:57561 ε3 g
− − − reducing the error by a factor 2 as compared with the N1st = 3 case.
Q = 4 + 10e T + 6e T + G3 e T ð50Þ These errors are practically independent of the lowering of the
ionization potential used to truncate the partition function as can be
Fig. 13. Exact values of nitrogen thermodynamic functions (Q, ε [eV], C = v) versus Fig. 14. Exact values of oxygen thermodynamic functions (Q, ε [eV], Cv) versus
temperature, calculated considering Δs = 500 cm− 1. On the right axis, 1st and 2nd order temperature, calculated considering Δs = 500 cm− 1. On the right axis, 1st and 2nd order
percent error have been reported. percent error have been reported.
872 G. Colonna, M. Capitelli / Spectrochimica Acta Part B 64 (2009) 863–873
Table 4 (ground state + lumped level describing excited states) can be used
Maximum of the relative error calculated with respect to temperature and number of with a fair amount of confidence. The maximum of first order error for
levels in the three-group model, for N and O thermodynamic quantities.
atomic hydrogen decrease by about 1% to about 0.1% for partition
Nitrogen Oxygen functions which increase the number of excited states from the
⁎ 1 2 ⁎ 1 2 principal quantum number n = 3 to n = 40. Similar deviations are
Q 3.1 3.0 0.49 2.1 2.2 0.68 observed for the internal energy and the specific heat. Second order
U [eV] 2.4 2.2 0.49 1.5 1.0 0.71 errors are always much lower than the corresponding first order
Cv 5.0 4.6 1.10 3.3 2.1 1.40 errors. The situation changes for atomic systems which present low
⁎: Approximate error (see Eqs. (31)–(41)); 1: first order error; 2: second order error. lying excited states either as a rearrangement of valence electrons (N,
O) or originating from L-S coupling in the ground state (O). These
appreciated in Table 3, where we have reported the maximum first levels should be taken into account during the reduction procedure.
and second order approximation errors found for the different Δ The final step consists in reducing these systems to a three level
values (from 250 to 1000 cm− 1). model, one characterizing the ground state, one the low lying excited
The situation for oxygen is different due to the level structure of states and the last one the lumped level accounting for the very large
this species, as can be observed from Table 1: in fact the ground state number of levels characterizing the given system. The advantage of
of atomic oxygen is split into three levels due to the L–S coupling, such a procedure is the possibility to introduce a self consistent
which have close energies. Then we have two low lying excited states thermodynamic model for extreme conditions characterized by large
(1D and 1S in Table 1) originating from the rearrangement of p4 levels. variations of electron densities i.e. with different cutoff values. In such
After that, the very large number of high lying levels originates from cases use of Eqs. (53) and (54), taken as an example, in combination
the interaction of the more stable oxygen core 4S with the excited with the parameters reported in Table 2 can strongly reduce the
electron jumping on ns,np,nd,nf… levels. A complete database for computational times allowing also a self consistent thermodynamic
atomic oxygen including observed and unobserved levels has been model in the fluidynamics equations. Future work in this direction
compiled by our group (see Ref. [14]) and directly used in different should try to reduce the existing databases to three level systems as
papers [15,29,34]. This database has been reduced in the form of well as to complete existing databases on observed levels (e.g. Moore
Table 2 for different cutoff values, to trasform the database in a single [20] , NIST [32], IAEA [33]) to transform them in few level systems.
lumped level with given G and ε for i N 5. This new database can be used with a fair amount of confidence for a
g
The results show that the selection of N1st should take into account self-consistent description of thermodynamic and spectroscopical
as separated levels those near to the ground state and the two low properties of LTE plasmas avoiding the complications of calculating
lying excited states. Inspection of Table 3 explains this point. We can in the electronic partition functions through a level to level approach.
g
fact appreciate that the choice of N1st = 1 as expected, gives strong
g
errors. The errors persist up to N1st = 5 when maximum first and
second errors become really acceptable in all the temperature and Acknowledgements
cutoff ranges. These results demonstrate that partition function and
thermodynamic properties of atomic species can be reduced to a few This work has been partially supported by MIUR under contract
level system grouping all the high lying excited states in a single PRIN 2007H9S8SW_003 project and Progetto Strategico Regione
lumped level. Puglia PS-136 “Sviluppo di un rivelatore a film di diamante per
A further reduction can be operated by grouping also the low lying radiazione ultravioletta".
excited states thus reducing nitrogen and oxygen to a three level
system. We can in fact write the three level partition function as References
properties of high–temperature mars–atmosphere species from 50 k to 50000 k, [25] O. Cardona, E. Simonneau, L. Crivellari, Analytic partition function for plasmas, Rev.
Esa str-246, isbn: 92–9092-349–0, European Space Agency (2005). Mex. Fís. 51 (2005) 476–481.
[15] A. Casavola, G. Colonna, M. Capitelli, Modeling laser-induced plasma expansion under [26] M.R. Zaghloul, Reduced formulation and efficient algorithm for the determination
equilibrium conditions, J. Thermophys. Heat Transfer 22 (2008) 407–413. of equilibrium composition and partition functions of ideal and nonideal complex
[16] M. Capitelli, A. Casavola, G. Colonna, A.D. Giacomoa, Laser-induced plasma expansion: plasma mixtures, Phys. Rev. E 69 (2004) 026702.
theoretical and experimental aspects, Spectrochim. Acta Part B 59 (2004) 271–289. [27] G. Colonna, A. D'Angola, A hierarchical approach for fast and accurate equilibrium
[17] H.R. Griem, High-density corrections in plasma spectroscopy, Phys. Rev. 128 (1962) calculation, Comput. Phys. Commun. 163 (2004) 177–190.
997–1003. [28] G. Colonna, Improvements of hierarchical algorithm for equilibrium calculation,
[18] H.R. Griem, Principles of Plasma Spectroscopy, Cambridge University Press, Cam- Comput. Phys. Commun. 177 (2007) 493–499.
bridge, 1997. [29] A. D'Angola, G. Colonna, C. Gorse, M. Capitelli, Thermodynamic and transport
[19] A. Bar-Shalom, J. Oreg, W.H. Goldstein, D. Shvarts, A. Zigler, Super-transition- properties in equilibrium air plasmas in a wide pressure and temperature range,
arrays: a model for the spectral analysis of hot, dense plasma, Phys. Rev. A 40 Eur. Phys. J. D 46 (2008) 129–150, doi:10.1140/epjd/e2007-00305-4.
(1989) 3183–3193. [30] J. Aubreton, M.F. Elchinger, V.R.P. Fauchais, P. André, Thermodynamic and
[20] C.E. Moore, Selected tables of atomic spectra, National Bureau of Standards, Jun. transport properties of a ternary ar–h2–he mixture out of equilibrium up to
1949 Nbs-467. 30 000k at atmospheric pressure, J. Phys. D Appl. Phys. 37 (2004) 2232–2246.
[21] K.M. Roussel, R.F. O'Connell, Variational solution of schrödinger equation for the [31] M. Capitelli, E. Ficocelli, The contribution of electronic excitation to the total
static screened coulomb potential, Phys. Rev. A 9 (1974) 52–56. specific heats of high-temperature gases: a misinterpreted absence, J. Plasma Phys.
[22] Y.P. Varshni, Scaling relation for the energy levels of a hydrogen atom at high 5 (1971) 115–121.
pressures, Z. Naturforsch. 57a (2002) 915–918. [32] http://www.nist.gov/srd/index.htm[online] (2009).
[23] M. Capitelli, D. Giordano, G. Colonna, The role of debye-hückel electronic energy [33] http://www-amdis.iaea.org/aladdin/[online] (2009).
levels on the thermodynamic properties of hydrogen plasmas including isentropic [34] M. Capitelli, G. Colonna, D. Giordano, L. Marraffa, A. Casavola, P. Minelli, D. Pagano,
coefficients, Phys. Plasmas 15 (2008) 082115. L. Pietanza, F. Taccogna, High-temperature thermodynamic properties of mars-
[24] M.W. Chase Jr., NIST-JANAF Thermochemical Tables, American Institue of Physics, atmosphere components, J. Spacecr. Rockets 42 (2005) 980–989.
Woodbury, New York, 1998.