numerical calculations in a problem with heat conduction ... · abstract. the temperature u of an...
TRANSCRIPT
Numerical calculations in a problem with heat conduction andheat productionCitation for published version (APA):Geldrop - van Eijk, van, H. P. J., van Ginneken, C. J. J. M., & Gelder, van, D. W. (1973). Numerical calculationsin a problem with heat conduction and heat production. (EUT report. WSK, Dept. of Mathematics and ComputingScience; Vol. 73-WSK-05). Technische Hogeschool Eindhoven.
Document status and date:Published: 01/01/1973
Document Version:Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can beimportant differences between the submitted version and the official published version of record. Peopleinterested in the research are advised to contact the author for the final version of the publication, or visit theDOI to the publisher's website.• The final author version and the galley proof are versions of the publication after peer review.• The final published version features the final layout of the paper including the volume, issue and pagenumbers.Link to publication
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, pleasefollow below link for the End User Agreement:www.tue.nl/taverne
Take down policyIf you believe that this document breaches copyright please contact us at:[email protected] details and we will investigate your claim.
Download date: 23. Nov. 2020
Abstract
. The temperature u of an exothermic reacting chemical in a vessel, governed
by the equation
uu = /iu + Ae ,t
is studied. With the initial condition u(~,O) = uo and the boundary condi
tion u(~,t) = 0 at the wall of the vessel, a critical initial value u~ is
calculated such that if Uo < u~, the temperature remains bounded and other
wise, if Uo > u~, an explosion occurs. The cases that the vessel is a slab,
an infinite cylinder, or a sphere are considered. The possible steady
states, together with the questions of their stability are studied; this
study should be considered as a personal presentation of some results, which,
in principle, can be extracted from the literature.
The problem is solved numerically, using standard techniques for solving
partial differential equations (Crank Nicolson).
A number of numerical results are presented in order to justify the supposi
tions that are fundamental for the algorithm.
AMS Subject Classification (1970): 6SPOS
- I -
O. Statement of the problem
A vessel is filled with a chemical having the homogeneous temperature uO'
An exothermic reaction takes place. The wall of the vessel is kept at con
stant temperature. Therefore the temperature of the reacting mass changes
by heat conduction to the wall and by heat production due to the reaction.
The problem is to determine a critical value u~ such that, if uo < u~' the
temperature of the chemical remains bounded and that, if Uo > uO' the tem
perature tends to infinity (explosion).
The cases that the vessel is a slab, an infinite cylinder, or a sphere are
considered.
I. Mathematical model
The temperature of the reacting mass is governed by the equation ([1, ch. 2J)
ut = div(k grad u) + W(u)/(pCv) ,
where
t time
k coefficient of thermal conductivity
p density
Cv specific heat at constant volume
W heat production per unit time and per unit volume.
The heat production term W has the following form (Arrhenius' law)
where
( I • I)
(I.2)
E internal energy
R gas constant
AO constant depending on the reacting mass.
The linearization of (1.2) around uo' supposing that k is a constant and
choosing suitable dimensionless quantities, transforms (1.1) into
ut = t.u + Ae u • ( 1• 3)
- 2 -
Hence, the problem can be formulated as follows. Let V be the reactant and
let S denote the wall. Let u be the solution of
u> a
)ut = l\u + Ae , x EO V, t-u(=.,O) = uo , X EO V ( I. /f)-u(=., t) = a , x EO S, t > a •-
. * . * . . .Determ1ne Uo such that, 1f Uo < uo' u(x,t) rema1ns bounded; otherw1se, 1f
uo > u;, u(x,t) tends to infinity.
If V is a slab, an infinite cylinder, or a sphere, (1.4) transforms (after
appropriate scaling and because of symmetry) into
ut = u + ~ u + Aeu a < x < I ,t > a )xx x x
u(x,O) = Uo 0 < x < 1
u(I,t) = 0, u (O,t) =0, t > a ,x
(I.5)
where n = a if v is a slab, n = 1 if V is a cylinder, n = 2 if V 1S a sphere.
2. Some properties of the model
In this section some - for our purpose essential - properties of the model
will be discussed.
In principle, one can extract these properties from the li terature([ I,
ch. 3 + references], [2J).
The following should be considered as a result of the study of the litera
ture, with a personal presentation.
2. I. First, we study the steady state solutions of equation (1.5).
These satisfy
d2v n dv v--+--+ Ae = 0,dx2 x dx
dVI = adx x=O
v(l) = a .
a < x < 1)
(2. 1)
- 'J -
We write the solutions of (2.1). if any, as
vex) = v(O) + fey) •
where
y = lAev(O) /2 x •
It follows that f satisfies
d2
f + ~ .!!i + e f = 0dy2 y dy
dfl = 0dy y=O
f(O) = 0 .
(2.2)
(2.3)
The relation between v(O) and A fol10ws from the boundary condition for
x =
(2.4)
Lemma. (2.4). considered as equation in A with v(O) fixed, has exactly one
solution.
Proof. Multiply (2.3) by yn to find
or
n d2
f n-] df + n fy dy2 + ny dy y e = o ,
d (n df) =dy Y dy
n f-y e
Since :;!y=O = O. we obtain
dfdy = n
y
dfWe conclude that dy < 0, so f is decreasing and, clearly. lim fey) = -~.
y-+co
Thus. with v(O) fixed, there is exactly one vaLue of A satisfying (2.4).0
- 4 -
Conversely. considering (2.4) as equation 1n v(O) with A fixed. it turns
out that (2.4) has one or more solutions if A ~ A and no solution ifcrA > A • We now develop this assertion for n = O. 1 or 2.cr
(i) n = o.The solution f of (2.3) is given by
f(y) = -2 log(cosh(y/I:2» •
Hence. (2.4) reduces to
v'1' cosh (S) = S ,
(2.5)
"' fA v(O) /2where S =~- e •2 . 2
Consequently. A = 2/ (sinh (S ». where S satisfies S tanh(S) = 1.cr cr crNumerical values:
A ~ 0.88. v(O) ~ 1.19 •cr cr
The dependence of A on v(O) is illustrated in the following diagram.
A
0.5
0=-----.--'-4.---2....---------r
3---.....4---....S------r
6--- v(0)
V(O)cr: 1.19
(ii) n = J.
The solution f of (2.3) is now given by
2f(y) = -2 log(l + y /8) •
Using (2.6), (2.4) can be written as
8S = A(1 + S) 2 ,
where S - A v(O)- 8' e •
(2.6)
- 5 -
Consequently, A = 2, v(O) = log 4 ~ 1.39 •ct' cr
The dependence of A on v(O) is illustrated in the following diagram.
A
1.5
o,s
'-----r---+----r---~---....,.....---...,......-'----........,-_ V (0)o 1 2
V(O)cr: 1.393 4 5 6
iii) n = 2.
In this case the solution of (2.3) is not analytically known. From
numerical calculations it follows that A ~ 3.32, v(O) ~ 1.61 and, cr crthe following diagram.
A
2
o 2 4v (Q)cr: 1.61
6 8 10 12 14
- 6 -
2.2. We are now going to investigate how a small perturbation of a steady state
solution varies with time.
Let
u(x,t) = vex) + ~(x,t)
be a solution of
u = u + ~ u + Aeu
}t xx x x
u (O,t) = 0, u(I,t) = ° ,x
(2. 7)
(2.8)
where vex) is a steady state solution and ~ a small perturbation of it.
Substituting (2.7) in (2.8) and linearizing we obtain
CPt = CPxx + i ~x + Aev(x)~ = OJ
cP (O,t) = 0, cp(I,t) = 0 .x
(2.9)
We look for non-trivial solutions of (2.9) by applying the technique of
.separation of variab les.
Supposing that
cp(x,t) = F(x).G(t)
we find
dG -Atdt = -AG, so G(t) = e
2~ + ~ ~ + (Aev(x) + A)F = 0dx2 x dx
(2.10)
(2. I I)
dFI - 0dx x=O - , F(l) = 0 ,
where A is a constant such that (2. II) has solutions that are not identical
ly zero. A is called an eigenvalue and F(x) the eigenfunction belonging to
it.
From the theory of Sturm-Liouville we have
a) The eigenvalues are real and can be numbered such that
with A -+ 00.n
- 7 -
b) The eigenfunction "'i (x) belonging to Ai has exactly i zeros in (0,1).
c) The [unctions 10' (x) form a complete set, so rp(x,t) can be written asn
00
<p(x,t) = Ln=O
-A tna F (x)e
n n
Consequently, if 1..0
> 0, the perturbation cp decreases (v(x) is stable);
otherwise, if 1..0 < 0, cp increases (v(x) is unstable).
Therefore, to investigate whether vex) is stable or unstable we try to
determine the sign of 1..0
,
For this purpose we look at the following problem that arises if we drop
the condition F(I) = 0 in (2.11) and put A = O. Hence,
d2F n dF ( )__ + __ + Aev x F = 0di x dx
(2. 12)
*Ix=o = 0 •
The solutions of (2.12) can be written as F(x) = CH(x), with H(O) = I and C
a constant.
Let E: be the first zero of H(x). Then it follows from the properties a) and
b) that zero is the smallest eigenvalue of
(Aev(x) + A)F = 0
:~Ix=o = 0, F(E:) = 0 •
Therefore,
if ~ > I, then(2.11)has only positive eigenvalues;
- if ~ < I, then(2.II)has at least one negative eigenvalue.
This follows from the fact that, according to the theory of Sturm-Liouville,
the smallest eigenvalue of
d2F n dF v( )- + - - + (Ae x + A)F = 0dx2 x dx
dFI = 0 F(a) = 0 ,dx x=O '
15 a decreasing function of a.
- H .-
o Summarizing, the problem of investigating whether v(x) is stable or un
stable reduces to the location of the first zero ~ from the solutions of
*Ix=o =0 ,
if t; > 1 , then v(x) is stable,
if t; < I , then v(x) is unstable.
The solutions of (2.13) are
F(x) = C(x :: + 2) ,
where C is a constant.
From (2.2) it follows that
dfF(x) = C(y dy + 2) ,
where y = ~ev(0)/2x.
For the first zero ~ of F(x) we have
~ = lAev (0) /2 '
where YO is the smallest root of
df + 2 0Y dy =.
(2. 13)
o
(2.14)
(2.15)
(2.16)
The function f is decreasing so it follows from (2.4) h IiJ v (0)/2 is ant at e
increasing and t; in (2.15) is a decreasing function of v(O).* of v(O) = 1, then it follows thatLet v (0) be the value such that t;
if v(O) < v*(O) then v(x) 1.5 stable,
-if v(O) > v* (0) then v(x) is unstable.
* is determined fromv (0)
- 9 -
or
*
·1v (0) = -f(yO)
and (2.17)** 2/ v (0)A YO e
Finally we show that A, considered as a function of v(O) given by (2.4),
*has its first extreme value in v (0).
From (2.4) it follows that the extreme values of A considered as a function
of v(O) satisfy
+ 2 = 0 •lAev(0)/2
f(lAev (0)/2) = -v{O)
lAev(0)/2 ~Idy .
y
(2.4)
(2. 18)
So we find lAev (0)/2 =tain
y., where y.~ ~
is a root of (2.16). Using (2.4) we ob-
v(O) = -f(y.) •~
The smallest value, v(O) ,introduced ~n section 2.1, where A is extreme,crequals
v(O) = -f(yO)cr*v (0) •
2.3. We conclude by making some suppositions that are useful for the algorithm
which we shall describe in the next section.
Only those cases are examined where an unstable steady state solution exists.
Let u(x,t,uO) denote the solution of (I.S). We suppose that, if at some time
to u(x,to'uO
) >vt(x) (0 < x < I), then U o > u~; otherwise if at some time
to u(x,to'uO) <vt (x) (0 < x < I), then Uo < u~.
Indeed, in all cases that were examined there was a time to where either
u(x,to'uO) >vt(x) or u(x,to'uO) <vt(x).
With this supposition u; can be determined by successive halving the inter
val (O,vt(O».
- 10 -
3. Numerical method
To solve the problem numerically, we start by replacing (1.5) by a finite
difference approximation (Crank Nicolson).
We introduce a rectilinear grid with sides parallel to the x- and t-axes,
hand k being the grid spacings in the
x- and t-directions respectively.
The grid points are:
t
1 lk
_1---+_-1-_1x·
1.1. * h, 1. O. I J •• • ,M, Mh =
t. =j *k, J =0,1 •••••J _h-l -x·
We write down the differential equation of (1.5) for the points
x. = i *h. i = O,I •••••M-l1.
tj+~ = (j + Dk, j = 0,1, •••
We have, denoting f (x. ,t.) by f. .,1. J 1.,]
u .. I - u. . 2( ) 1. ,] + 1. , J + 0 (k ) ,u t . . I = k1.,J+2
k t 0 (3. I)
u. . - 2u. . + u. . u. 1 . I - 2u. .+ 1= ~[ 1.+1.J 1.,J 1.-I,J + 1.+ ,J+ 1.,J
h2
h2
+ u. .1.-1 ,]+1] +
~f i :f 0
(3.2)
u. . - u. l' u. 1 . 1 - u. 1 '+1n I[ 1.+1,J 1.- .J + 1+ .J+ 1.- ,J ] += Ih 2 2h 2h
(3.3)
If i 0, then we have, because u (O.t) = 0,x
(3.4)
- II -
Furthermore,
It follows from the boundary conditions that
= 0
(u) . = 0 = u I , j - u-I , j + 0 (h2) h '" 0 • Jx O,J 2h '
(3.5)
(3.6)
If the order terms are neglected the finite difference approximation of
(1.5) arises by substituting (3. I) to (3.5) incl. in the differential
equation of (1.5) for the points
U I . and uM . are eliminated by means of (3.6) •. - ,J ,]We formulate the above with matrices as follows.
We introduce the M-dimensional vectors
j = 0,1, ...
f (U.)- -J
kA
the tridiagonal M x M-matrices Band C
- 12 -
B :=
where
Uo . kaO := 1 + A(n + I) - ~Ake ,], A -
- h Z
U.•a. := I + A - ~Ake ~,J, i = I, ••• ,M-I~
nSo := -A(n + I), Si = -~A(I + Zi)' i = 1•••••M-2
Yi = -P.. (I - ~i)' i = 1, ••• ,M-1 •
The matrix C is found from B by replacing h by -h.
As an approximation of u.. in (1.5) we take U.. which satisfies~,J ~,J
BU. 1 = CD. + f (U . ), j = 0,1, •••-J+ -J - -J
(3.7)
UoGiven U. we can calculate U. 1 from (3.7) solving the tridiagonal-system.-J -J+If we look at the approximation in a fixed point (x, t), we expect, because
of (3.1) - (3.7), that
Z ZU(x,t) = u(x,t) + CI (x,t)h + CZ(x,t)k +
Z Z+ o(h ) + o(k ), h ~ 0, k ~ 0 • (3.8)
We make the following suppositions (cf. section 2.3), which in all cases
examined appeared to be satisfied.
- 13 -
(i) The difference equation ~n (3.7) has (with appropriate A and h) at
least two steady state solutions, a stable one US and an unstable oneR.
U •
(ii) Successive solving of (3.7) with uO' h, k and A given, we find a jo
h . < UR. R.suc that elther U. or U. > U •-J O -J O
*Let UO(h,k,A) be defined so that, if Uo(3.7) successively, a U. <_uR., and if
-Jtively. 0
uri (h,k,A) is taken as an approximation of u~.
find when solving
U.. > UR. respec-J O
*then it follows that either Uo > UO(h,k,A) or
*Now UO(h,k,A) ~an be calculated by successive halving of the intervalR.
I := (O,(U )0)'
*Given h, k and A, we use the following algorithm to calculate UO(h,k,A).
I) Calculate UR..
2) Determine uo by halving I.
3) Solve (3.7) successively;
*Uo < UO(h,k,A).
4) Adapt I and go back to 2.
Sub 1. To calculate Ui we proceed as follows:
A steady state solution of (3.7) satisfies
(B - C)E. = ! (.£) •
To solve this non-linear system we use Newton's method
a,s.(E - - )Un+ 1
aUu=un -
where
a~ .n £. n= ~(U ) - (-- )U , n =£. - au n -u=u.... -
(3.9)
E = t (B - C) ,
- 14 -
If QO is appropriate, we have lim ~n = ~£. As QO
we take the unstablen-HlO
steady state solution of the differential equation in (1.5). If n = 0
or n = I, then this solution is analytically known. If n = 2, we deter
mine it numerically.
We expect that, analogously to (3.8),
* * 2 2 2 2UO(h,k,A) = uo + C1h + C2k + o(h ) + o(k ), h '" 0, k '" O. (3.10)
Numerical experimentation has supported this supposition and thus we can
apply extrapolation, if desired so.
4. Numerical results
As an illustration of the numerical method we present some intermediate re
sults for the case n = 0, A = 0.5.
Steady state solutions of the difference equation (3.7).
h = 1/8.
Stable solution US Unstable l· £x. so ut10n U1
0 0.329 301 216 8 2.891 333 774
o. 125 0.323 871 543 1 2.820 953 600
0.25 0.307 641 324 7 2.619 379 233
0.375 0.280 784 442 2.310 561 119
0.5 0.243 582 495 9 1.922 992 386
0.625 o. 196 413 271 8 1.481 975 217
0.75 0.139 736 002 8 1.006 570 302
0.875 0.074 074 591 7 0.509 788 821
0 0
- 15 -
I ISolving (3.7) successively, h = 8 • k = 64 •
We find with Uo = 2.272 672 213 04
2.889 856 0
2.819 574 4
2.618 258 3
2.309 770 7
!!30 = I. 922 516 7
I. 481 740 8
1.006 483 6
0.509 768 6
0
and with U o = 2.273 705 115 69 we find
!!21 =
2.892 422 8
2.822 482
2.622 006
2.314 421 5
1.927.668 0
1.486 669 I
1.010 365 8
0.5119070
*APproximations of uo•
k = h2
*h Uo(h ,k .A)
1/4 2.28612
1/6 2.27592
1/8 2.27283
1/ 10 2.27154
1/12 2.27089
1/16 2.27029
Extrapolated values
2.26776
2.26886
2.26925
2.26941
2.26952
- 16 -
k = h
*h Uo (h,k ,A)
1/6 2.32287
1/8 2.30475
1/12 2.28844
1/16 2.28145
Extrapolated values
2.28145
2.27539
2.27246
To choose a value of k, when h is given, we have to take into account a
(3. 10).
hand, it appears from the table that a
of uO• This means that Ic21 » Ic] I in
IC1 IAn optimum value of k is about rc;T h.
I c2 1In this case we have~ - 10.
number of contradictory demands. On the one hand we want a large k to reach
quickly the time to' whereupon either U. > Ui or u. < Ui. On the other
-J O -J Olarge k leads to a bad approximation
* 1As an illustration we tabulate the approximations uo for k = 3 h.
*h Uo(h,k ,A)
1/4 2.29147
1/6 2.28146
1/8 2.27702
1/12 2.27331
1/16 2.27184
Extrapolated values
2.27345
2.27131
2.27034
2.26995
For practical reasons we have dropped the idea of trying to determine an op2timum value of k in all cases. We always took k = h •
*Finally, we present a table and a plot of uO• The tabulated values were ob-
tained by extrapolation and have a relative error of at most approx. 0.001.
Our algorithm was not applicable to small values of A for n = 2. These cases
were not further examined because there was no need for them.
- 17 -
*Table of uo·
* *n = 0 A Uo n = 1 A UoO. 1 4.30 0.3 3.726
0.2 3.48 0.5 3.143
0.3 2.976 0.7 2.742
0.4 2.591 0.9 2.429
0.5 2.270 1• 1 2. 162
0.6 1.979 1.3 1.923
0.7 1.700 1.5 1. 697
0.8 1. 395 1.7 1.467
0.85 1. 197 I .9 1. 194
n = 2 *A Uo1.8 2.039
1.9 1. 970
2.0 1.900
2.2 1.771
2.4 1.645
2.5 1.583
2.6 1.520
2.8 1. 392
3.0 1.254
3.2 1.082
3.25 1.025
- 18 -
*U
f4,8
4,0
3.2
2.4
1.6
.8n: 0
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1.0
-A
*U
f4.0
3.2
2.4
1.6
.8
n : 1
0 .2 .4 .6 .8 to 1.2 1.4 1.6 1.8 2.0
-A
*Uo
t 2.0
1.8
1.6
1.4
1.2
1,0
1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2
----- A
- 19 -
References
[IJ Tipper, C.F.H.
Oxidation and Combustion Reviews. Vol. 2.
Elseviers Publ. Comp., Amsterdam, 1967.
[2J Istratov, A.G. and Librovich, V.B.
On the stability of the solutions in the steady theory of a
thermal explosion.
Prikl. Mat. Meh. 27 (1963), 343-347 (Russian).
Translated as J. Appl. Math. Mech. ~ (1963), 504-512.