Sunteți pe pagina 1din 4

The Difficult PDE Files

Part 1 : The Heat Decay Equation

Welcome to my collection of LATEX files that deal with the more unlcear problems and ideas that
arise in Partial Differential Equations this semester. I hope that these files will help others learn
this difficult mathematics class as it helps me organize my thoughts and makes the subject matter
clearer.

Our first topic, The Heat Decay Integral Problem

The first idea that we visit is the one of the first difficult siutations in PDEs. This problem is
the heat decay problem. The best way to see this problem at first is to understand it’s qualitative
behavior before tackling the arduous calculations involved. In calculus III, we learned of multivariate
functions that return planes as their result rather than lines and curves (this is where the idea of
a curve as degenerate plane comes from). The ”heat decay” problem describes a certain PDE
called the diffusion equation and this equation desribes the dynamic behavior the function plane
mentioned above.

Whenever you see examples of that diffusion process or any dynamic process, a PDE describes
how a function plane result changes with time and how function values on that plane vary over
space (Note this is just for the f2 → f1 case, but it’s a good place to visualize the process). In the
diffusion process, the distribution of function values ”spreads out” over time on that plane, and
this process takes place over certain region that the function covers called the domain. This can
best be seen through the heat conduction problems that we’ve seen in class, and through many
physical perspectives. One of those examples is when you see an iron schmelter heat the end of a
metal bar so that it glows red hot. The way the red hot metal bar cools to regular temperature
from the flame to the other end as a gradient, and how the red hot part slowly cools as the flame
is removed is a great example of the diffusion process in terms of heat energy.

Initial Conditions, Boundary Conditions, and the idea of arbitration

How the values of the function ”spread out” depend on multiple details. The best way to see
these details is through the equation that we were given on this homework problem. That equation
and it’s associated details are below :

ut = kuxx , u(0, t) = u(l, t) = 0,


and u(x, 0) = uo (x) for 0 ≤ x ≤ 1 and t ≥ 0.

Here, the several details needed in order to solve this problem are given. For the sake of simplicity
the unit value of one is chosen as the domain that this PDE covers. This fact is given from the
x inequality. Also this PDE is of one dimension and that is why x is the only spatial variable in
the problem (if there was a y element, then there would be a uyy somewhere in the PDE). The
u(0, t) = u(l, t) = 0 values are called boundary conditions and define what the function values
should be at the ends of the domain. This is where the details in this problem get a little tricky.
You can identify boundary conditions by seeing that the variable t in the u functions is not a
fixed value and remains symbolic, and that the x spatial variable is defined at length 0 (ie u(0, t)).
However the next function in the equality (u(l, t)) has no fixed valuefor x and instead contains an
l. How can this be done ? The answer is an idea in mathematics called arbitration where a variable
can stand for an arbitary value. This implies that in our problem that the length of the domainin
our problem can be of any value and this PDE model is still valid. Though not always obvious,
arbitrary variables are common in mathematics and are essential to understand what is happening
in certain situations. This idea of arbitration also arises in the next detail of the problem called
the initial conditions. In PDEs, the function that describesthe field at the first time step (t = 0)
is called the initial conditon(s). This is similar to ODEs and their initial values, but instead what
is given is a function. In our problem the initial condition is set to an arbitrary function that is
f 1 → f 1 with the x spatial variable as its single input.

Finally we arrive at the actual PDE itself, ut = kuxx . The answer to a PDE is a function (if
simple enough) that contains the variables t and x. This solution would describe how the function
value distribution would change over time. This symbolic notation for the solution is u(x, t) and
that is how the intial and boundary conditions recieve their notation. We are in fact defining parts
of the solution itself in that part of the problem. The k coefficient is a constant called the diffusivity
and is a physical property of the system that we are modeling. For simplicity, the diffusivity is set
to the unit value of one as well.

The integration relationship proof

Now that the PDE model has been analyzed, we can now begin to prove the integration rela-
tionship that was presented in the homework problem. The relationship that was given is defined
below :

Rl Rl
0 u(x, t)2 dx ≤ 0 uo (x)2 dx, ∀t > 0

At first the meaning of this inequality seems a little vague, but now that the PDE model is
analyzed thoroughly, the relationship is revealed. The u(x, t) term is recognizable as the time
dependent (or dynamic) function solution to the PDE explained before. The uo (x) term is the
initial condition given as an arbitrary function and doesn’t change with time (non-dyanmic). Since
we now know how the decay process qualitatively behaves and what the functions are in the
integration inequality, it is clear to see that the area under the initial condition curve is always
greater than the dynamic answer to the PDE. Think of the u(x, t) function as a normal distribution
in one dimension and as time progresses the curve decays to a flat line along the x axis (this is also
because of the fixed boundary conditions at 0). Therefore no matter what the time, the decayed
version of the original curve always has less area under it than its intial counterpart. (Proof note :
”∀t” means ”for all values of time”)
Now the really tedious symbolic proof calculations can begin. First define a f 2 → f 1 function
(E(x, t) in the hint) that has the same spatial and time inputs as our PDE answer to equal that
answer’s integral from the relation above :
Rl
E(x, t) = 0 u(x, t)2 dx

From here we take the derivative in respect to time of the integral, which seems a little strange.
However, when you think of the curve decaying to the straight x axis line in our qualitative
perspective it decays at a certain rate and since time derivatives are rates of change then the
reasoning for the differentiation makes sense :

d d
Rl
dt E(x, t) = dt 0 u(x, t)2 dx

Now we can apply the ”Flippy Floppy Theorem” to drag the differential inside the integral,
however there are some exceptions in this particular situation :

d
Rl d 2
dt E(x, t) = 0 dt u(x, t) dx

Since the answer to the PDE u(x, t) is squared, then you cannot simply make the squared
function a partial derivative of time (u(x, t)t ) like before. Now we must use the regular chain rule
symbolically to effectively apply the ”flippy floppy theorem” :

Rl d 2
Rl
0 dt u(x, t) dx = 0 2u(x, t)u(x, t)t dx

Here we subsitute the u(x, t)t term by the relation given in our original PDE, ut = kuxx . This
relationship doesn’t seem obvious at first, but ut is in fact u(x, t)t written in a simpler way. PDEs
are written by the derivatives of their function answers like ODEs and this subsitution uses this
fact :

Rl Rl R1
0 2u(x, t)u(x, t)t dx = 0 2u(x, t)ku(x, t)xx dx = 2k 0 u(x, t)u(x, t)xx dx

Now we apply integration by parts to split up the integral into terms that can be analyzed to
see whether they are greater, equal to, or less than 0. First we choose the two parts of our integral
(f (x) and g 0 (x) respectively) and apply the integration by parts formula from calculus II to get
our expanded result :

f (x) = u(x, t)
g 0 (x) = u(x, t)xx

Rb Rb
a f (x)g(x)dx = f (x)g(x)|ba − a f 0 (x)g(x)dx
Rl R1
⇒ 0 u(x, t)u(x, t)xx = u(x, t)u(x, t)x |l0 − 0 u(x, t)2x dx

Note that in the integral on the far right of this equation, the u(x, t)2 term arises from diffe-
rentiating u(x, t) in terms of x and antidifferentiating u(x, t)xx by the same variable. When that
calculation is performed 2 u(x, t)x terms arise and are mulitplied to get the square. Finally the eva-
luation term can also go to 0 because of the fact that u(x, t) = 0 along the entirety of the domain.
This is given from the boundary conditions (u(0, t) = u(l, t) = 0) and is from the assumption that
an intial value distribution throughout a domain is 0 when all its boundary conditions are 0. By
using this assumption we can reduce the integral even further :

R1 Rl
u(x, t)u(x, t)x |l0 − 0 u(x, t)2x = − 0 u(x, t)2x dx

This is the integral that we have been looking for ! Since the integrand u(x, t) is squared then the
integral is always positive and when multiplied by the negative one outside the integral sign then the
integral is always negative. This means that the rate of chane of our earlier dyanmic function E(x, t)
that we used to define our integral with is always negative. By our earlier qualitative perspective,
we have numerically proven that the solution curve decays with time and has its maximum dynamic
value at t = 0. Therefore the dynamic curve always has a smaller area than the initial curve from
uo (x, t). This is the perspective that is implied when :

Rl
E 0 (x, t) = − 0 u(x, t)2x dx ≤ 0

And therefore this is the final step of this lengthy and difficult proof !

S-ar putea să vă placă și