Interpreting the solution of the heat equation on the whole line

360 Views Asked by At

Can someone provided some insight on what is meant by "weighted average"?

Allow me to give some exposition.

So given the heat equation on the whole line

$$\begin{cases} u_{t}-ku_{xx}=0\quad\quad -\infty <x<\infty ,\;t>0\\ u(x,0)=\phi (x) \end{cases} $$

we know its solution is given by

$$u(x,t)=\int_{-\infty}^{\infty}S(x-y,\;t)\phi(y)\;dy,$$

where $S(x,t)$ is the heat kernel

$$S(x,t)=\frac{1}{\sqrt{4\pi kt}}\;e^{-x^2/4kt}\quad t>0.$$

The heat kernel, as we know it, represents the evolution of temperature for some region in space. In this case, for a 1-D infinite rod.

Now, in Walter A. Strauss' An Introduction to Partial Differential Equations, Strauss claims that the solution $u(x,t)$ is "a kind of weighted average of the initial values around the point $x$."

He later justifies this notion by writting the solution as follows

$$\int_{-\infty}^{\infty}S(x-y,\;t)\phi(y)\;dy\simeq \sum_t S(x-y_i\;t)\phi(y_i)\Delta y_i $$

and suggests that the solutions $S(x-y_i)$ are weighted by $\phi(y_i)$ .

My interpretation of the solution (in the context of 1-D heat flow in a rod) is that once time starts, the initial temperature distribution $\phi(x)$ spreads throughout the rod and as time goes on, the temperature value at a fixed point $x$ is given/determined by the average value the temperature of nearby points.

Is my intuition on track?

Also, what exactly does Strauss mean when he says $u$ is a "weighted average"?

Any input is much appreciated.

1

There are 1 best solutions below

1
On

The literal interpretation of this formula is as follows: The function $S(x-y_1)$ is the normal distribution shifted from $x=0$ to a specific $x$ value $= y_1$. All values of this shifted function along the $x$-axis is then multiplied by all the corresponding values along the $x$-axis of the (unshifted) initial condition function. This number is multiplied by delta $y$. All these numbers are added up. As the area below the normal function is $1$, the resulting sum becomes a weighed average, i.e. the initial condition is weighed by the corresponding values of the normal distribution located at the specific position $x=y_1$. To get the value at another $x$ position, you have to shift the normal function to that new $x$-position and repeat the calculation. The procedure may be described as finding a moving average or as a convolution.