O, ~20, independent of h=(O,H). Then the equation (1.2) has a unique solution v,,* and there exist the elements w,=E,, ( n, independent of h=(O, H), so that we have
j=
(1.12)
Note 4. The existence of vh'may be considered we want is the existence of (1.12). Proof. Since 'p is constructive solution vh' and I,$ -+uh* when 1-a. it is not difficult to verify that
as an extra
result,
the main conclusion
it is obvious that the equation u=q(u) has a unique It is also obvious that v,,' issolution of (1.2). No"
(I+*) IIIh. ll"h -"h(i'~),I~p~~"n("-":l-~' Hence
we have IIvh.-v:' II
From
(1.11) we have
Hence
(1.13) Now we choose n
(0) “A
h"wj. -Pihu'+p,n z I-=
u=u*. Then by Lemma 4 we have where w,,j-m,...,n, have been determined in Lemma 2 with ). On the other hand, by the assumptions about&and p we have llTa(vlp' )IJzh=Os, (h'"+'" Il&ll=G(h-'9, Then
(1.13) yields
and the theorem
is proved.
7. Theorem 4. Let independent of h=(O,H),
Assume
Eu=E,h. so that
that there exist constants
II[Z’h’(“h) 1-'/
to a neighbourhood
o,, of
M,p,g:
M>O, pao, q>o
IIT/(~h)llGfh-"
.plhu* which contains the element
the u:,=E,,j=?n,..., n, have been determined in the Lemma 2 with "=a'. Then for n large enough and h small enough the equation (1.2) has a unique solution and we have the expansion
where
Note
5.
Proof.
Anaogous
to Note 4.
we write down the Newton'smethod
for the equation
(‘+“=Uhtl)_[Thl(y~l’)]-iTh(“~‘)( Uh and choose
Then for
the first approximation
hE(O,H)
we have
as
(1.2) :
c,,'
53
Hence by Lenrna 4 we have A'&&(A-oP+~II
)ilr,(V:"'))lul=OH(h'"+*-rp-.").
So for n large enough to get ni-h-2p-q>O
and then for h small enough to have KBc'], the Kantorovitch's Theorem (Theorem 6 in /14, p.632/) is available, that is, the equation (1.2) has a unique solution v,,'in uh and ,]r;-r:D, Illl=Otl(h(l+b-'~-.)r). Therefore
of
... . ho):
the theorem
is proved.
2, Multiparameter formulae for operator equations. H
<
For a quantity cp(h)=cp(h,,...,h.) we write v(h)=0,(]]hll"), a being a real number, if there exists a constant M>O independent of h&8, so that (m(h)(
E***h'
lim Ilphull~=llu(l~, u=E,. lW-0
Assume yet that to each k(h) we can construct an application T,O:E,o-E,h being an approximation to T. By the abstract (multiparameter) finite difference method we want to say all the process of constructing the spaces Eo, the operators p,&the application Th. and replacing (1.1) by Tho(uh) =O, (2.1) W=E,h. This method is denoted by The. Assume that the equation (2.1) has a unique solution u*' for e.ach h=(h). Then the expression ekPnr'-p,,,n~~Elh is called the error of the method and ]]eh]lln the aboslute error. If (]el]],h+Owhen ]]A\]+0the method is called to be convergent. If there exists a positive number a so that ()e,Jl,p-OX(Ilhl~) the method is called to have an accuIf there exist two numbers r,h (~>O,O
c i: w,,...,.\\ ,h==OX(Ilhll’“+*~‘)
11 eh-P,hxz 4-m
(2.2)
,,+...+,a-i
the error is called to admit an asymptotic expansion with formulae. n. This expansion is of the multiparameter
respect
to o parameters
up to order
In order to simplify the writing of the formulae we introduce 2. Some abbreviations. some abbreviations. Let us denote by I the set of a-dimensional vectors with non-negative integral components, i.e. ifl if t-(i,, . . . . i.},i,-integer>O. If &I we write If] for denoting the sum of all components of i: Ii]++... +i. amd we write [i] for denoting the index i,...i, for instance mli,means w,, ,0. Now for &ml and h=a that is i={i,,...,i.),h--(h,,...,h.} we write h- for denoting the product A,'1 ...h >. So we have (h*')r-(h,'t...h.'~)~-h,"T...h~. With these abbreviations
the formula 11Ch-Pth2
(2.2) can be written
simply
as
(A")%,, 11,h=6dllhll’“+““).
c
I-n 111-i Henceforth 3. The family of starting methods. methods Th. having the following properties. Thesis 1'. There integers a,m (n>m>l),
Q!‘, lil -m..
of
he,%,
11 Tm(p,,u)-pu[
so that for
The is Frechet
a*-constants
a positive number and the operators
two Hcl, FI,,: a-+
we have
111-c
differentiable
and (2") denotes
UEQ,
rhO of the
T(u)+% r, W’)‘F,I,(~)]I)~ =~x(llhll’“+‘“). ‘-In
where
only the family
exist two real numbers r.h (y>O,Occk
, n, independent
Thesis 2'. FA,h we have
we consider
(s,...,s),k
up to order v with m(v+l)~n+l
and for uh,UIE
times.
A(u): D(A)cE,+E, Thesis 3'. There exists a linear operator a parameter and can be approximated by ThO'(u) on D(A), i.e. for
which depends on II-ED(A) we have
ue&,
as
54
il~~.‘(~,~u)~,~t~-p,A(u)c~Il,~-O when JI~(J-_o i=m,...,n, has a unique solution
and at the same time each equation n(n)~=g&,', with Q,'zQ:tl.
Thesis 4’. hdf,
so
There exist the operators that for II'%&,wES&', i-m,. .., n,
Cu,:Q,X~,'+62 :A{,Ii]-1.2, . . . . n-i, we have
u.%?,'cD(.A)
independent
of
II n-i
c c
(h@‘)‘G,,, (4 WI
111-t
i-1
Ill
Ih=OX(llhll(“-‘+~“),
G,,,=O.
Thesis 5'. There exist the operators r,+...+c+ul
* G,,,: Q,XR,'~X... k=2, 3,. . , v, Isj=m, .. . , n,
i,=m,. . . . n, ljl=O,
l,...,
n-km, izzzkt
of
'k%.
uq&,
so that for
w,.~EQ,“I,
we have II
T2 (p,rd (PthW [,,I, . . 1plh~I.,I 1n-m
cc
(ha’) q:, p-0 111-s
PM Note 6.
Analogous
(4 W[#lll. . . >w
-_O1r( Ilhll(‘-*m+L)’1.
to Note 1.
4. The results. we can prove four lemmas analogous to the Lemmas 1-4 and after that the following four theorems. we always assume that Th.~t, and u'EQ, is the unique solution of the equation (1.1).
Theorem 5.
Assume that 1) the equation (2.1) has a unique solution vh'; 2) II[T~,‘(P,*~‘)I-‘I~~M~-~’ where M and p are constants M>O, OdpGm, 31 the method Tha has an accuracy of order pi with q>p. Then there exist the elements wlr,'E,.]j]-m,..., n, independent of
independent hd,
so
that
of h&S
;
we have (2.3)
5.
Theorem 6. Assume that 1) as in 1) of Theorem 5; 2) there exist two constant c and p (c>O,pbO), independent of M, so that for vn+ we have u~=E,h I/T~.(~~)-T~.(~~)lI~~ch~'Ilu~--l(~~~,~. Then there exist the elements h=S so that we have Q,, IjI=m,.. . , n, hkpendent of the expansion (2.3). 6. Let B, be a linear operator: .l&-+E,, which depends on h&S, so that ~~B$GWrPT where M and p are constants, M>O,p>O, Then for a fixed hm we independent of hd8. consider the iterations ll+'r_yh(') B T (s;)) -hha . Vii UEE,,, is constructive for each Theurem 7. Assume that the operator cp(u)-u-BhTn.(u), hc%, i.e. for v,n~;B,~ we have ]]cp(u) where O
O, qZ0, independent of ha. Then the equation (2.1) has a unique solution II,,* and there exist the elements wu@,. so that we have ljl=m,. . . , n, independent of ha,
II e*-p,h2 r, UP9%, 11,~OX(Ilhll(~+L-~--p)T). t-7”Iii--i 7. Theorem 8. Let E,,,=EII. independent of hE.#, SO that
Assume
that there exist the constants
]][Th.'(~h)] -‘I(Gfh-P’, for any vk belonging
to a neighbourhood
oh of
M,p,
q,bO,p>O,q>O.
IIThs”(un)II
PlhU. which contains
cu,‘,=E,, ljl-m,. . . , n. Then for n large enough and (Ihl( small enough the equation WE' and we have the asymptotic expansion for the error:
the elements
where
(2.1) has a unique solution
55
of
wu, are independent
where
3.
hEa.
Some numerical examples.
In this paragraph Example
1.
we give some numerical
Consider
examples
for illustration.
the problem
Au=f(r, y), OCz
O
U(0, y)=u(i, y)--u(s,0)=u(z,1)=0,
Its solution
is
u(r, y)=sinnzsinny.
We use
z,=ih, h-l/N, k-l/M, y,=ik, where N, M are positive integers satisfying O
the grid
any discrete
I% Y,)
function
AhuI,=h-1(Ui+,,j-2u,,Su,-,,j) +k-‘(uii+,-2uij+vt.l-,). Consider
the discrete
problem
Ahvi,j=ft.hOO, k>O so that v(P; h, k)--U(zp, yp) =h’w,,o(zp,
YP)+
k’wo., (zp, YP) +h’wt.o (~3 YP) fh’f-*w,s
(G,
VP) +
k‘w,,t(zp, YP) + . . . +h*“w,o (zp, yp) +h’“-‘k*w._,,,
(zp, yp) +
..
k’“w,,.(zp, y,)+O(h;“+z+kz”+z), where n denotes
any positive
$
integer.
Form that we deduce
v(P; h, k) ==u(zP, y,)+O(h’+k’).
Then u(P;h, k;h/2,k/2) may be considered as the first ameliorated approximate value of u(fp,yp) The numerical results are presented in the Table 1 in which the last line is added only to help us to make some comparison of approximate values at p('/,,'/,).
I
ExamPle
2.
Consider
I
the problem
I
I
Table
1
/lo/:
3 yw'_ 9 Y',
o
y(O)=4,
y(l)-1.
Its solution is y(s)-4/(1+r)*. We use the grid (r,} with one parameter h: z,-ih, h-l/N, where is a positive integer. For any discrete function v defined over this grid we write 0, to denote ~(2,) . We consider the discrete problem
N
$(
3 u,+,-2u,+Uc-r)--IJ,*, 2
vr-4,
V&.=-i.
Resolving it we denote the approximate value calculated at a fixed grid point P of the grid with parameter h by u(P; h). Since here y(t) and its derivatives are smooth enough we can prove the existence of n continuous functions o,independent of h>O, so that u(p; h)--u(zr)-
where n is any positive y(rp):
u
etc.
The numerical
integer.
&P’q+O(h*‘+‘). I-I
From that we duced
the ameliorated
approximate
(P;h;S;f)-~u(P;~;a)-~” ( P;h;+)-y(zp)+O(h’) results
at the point
P(0.5)
are
presented
in the Table
2.
values
for
56 Table
Profiting by this occasion we expressourgreat thankfulness Zhidkov for precious advices and remarks, and effective help.
to N.N. Govorun
2
and E.P.
REFERENCES 1.
SAMARSKII A.A., The theory of difference schemes (Teoriya raznostnykh skhem), Nauka, Moscow, 1977. 2. MARCHUK G.I., Methods of computational mathematics (Metody vychiclitel'noi matematikil, Nauka, Moscow, 1977. 3. MARCHUK G.I. and SHAIDUROV V.V., Increasing the accuracy of the solution of difference schemes (Povyshenie tochnosti reshenii raznostnykh skhem), Nauka, Moscow, 1979. 4. AIRYAN E.A. and ZHIDKOV E.P., On one method for increasing the accuracy in the numerical solution of elliptic equations. Pll-12709, Joint Institutue of Nuclear Research (OIYAI), Dubna, 1979. Improving the accuracy of approximate solu5. ZHIDKOV E.P., NGUEN MONG and KHOROMSKII B.N., tions of a non-linear singular integral equation of the Chu-Lou type, PS-12916, Joint Institute of Nuclear Research (OIYAI), Dubna, 1979. 6. ZHIDKOV E.P., NGUEN MONG and KHOROMSKII B.N., Refinement of the approximate solutions of non-linear operator equations, P5-12979, Joint Institute of Nuclear Research (OIYAI), Dubna, 1979. 7. VOLKOV E.A., Investigation of one method for improving the accuracy of the method of nets in the solution of the Poisson equation, in: Computational mathematics (Vychislitel'naya matematika), Issue 1, 62-00, Izd. Akad. Nauk SSSR, 1957. 8. WIDLUND O.B., Some recent applications of asymptotic error expansions to finite difference schemes, Proc. Roy. Sot., A323, 1553, 167-177, 1971. 9. JOYCE D.C., Survey of extrapolation processes in numerical analysis, SIAM, Rev., 13, 4, 435-490, 1971. LAURENT P.J., &ude 10. des pro&d& d'dxtrapolation en analyse nume/rique, Thesis, University of Grenoble, France, 1964. 11. STETTER H.J., Asymptotic expansions for the error of discretization algorithms for nonlinear functional equations, Numer. Math. 7, 1, 18-31, 1965. 12. PEREYRA V., Iterated deferred corrections for non-linear operator equations, Numer. Math. 10, 4, 316-323, 1967. 13. TA VAN DINH Sur les formules asymptotiques de l'erreur en mgthodes aux diffe/rences finies, Acta I4ath. Vietnamica, 2, 1, 116-140, 1977 . 14. KANTOROVICH L.V. and AKILOV G.P., Functional analysis in mormalized spaces, (Funktsional'nyi analiz v normirovannykh prostranstvakh), Gostekhteorizd., Moscow, 1959. Translated
U.S.S.R. Comput.Maths.Math.Phys., Printed in Great Britain
Vo1.24,No.5,pp.
5662,1984
by E.L.S.
0041-5553/84 $10.00+0.00 01985 Pergamon Press Ltd.
APPROXIMATE AND NUMERICAL METHODS FOR CALCULATINGTHE COMPOSITION OF AN EQUILIBRIUM PLASMA* P.D. SHIRKOV
A simple mathematical model based on approximate separation of dissociation and ionization processes is proposed for calculating the particle densities in a plasma which is in local thermodynamic equilibrium in the region of dissociation and primary ionization of the atoms and molecules. Accurate solutions of this separated model are found and "a priori" A separated model is constructed estimates of its accuracy are obtained. Comparisons are for calculating the multiple ionization of a medium. made with numerical calculations of the composition of a plasma. An efficient iterative process is proposed for calculating the equilibrium densities of a dissociated and ionizing medium.
Introduction. 1. Calculations of chemical and ionization equilibria lie at the foundation of the calculation of thermodynamic functions and gas transfer coefficients and are exceedingly tedious *Zh.vychisl.Mat.mat.Fiz.,24,9,1372-1380,1984