Thread Subject: Create a vector of derivatives, given a vector of values

Am looking for recommendations about how to create a vector of derivatives given a vector of values.

In this case, I have an array with temperatures at different times (evenly spaced). I want to create an array of temperature derivatives. (I anticipate losing a few points at the ends.)

Could write a loop, I suppose, dT/dt_i = (T_i+1 - T_i)/dt
But there are 20,000 points.

Is there some vectorized scheme or existing function that I should look into?

Thanks.

"Johan " <robert.mchugh@ips.invensys.com> wrote in message <gd6dfl$be8$1@fred.mathworks.com>...

> Is there some vectorized scheme or existing function that I should look into?

help diff

Bruno

Thanks. "help diff" works much better than "help derivative." :)

Turns out the data is a little noisy, so estimating the derivative by comparing just two adjacent points produces oscillating derivatives.

Are there existing tools to smooth this out?

Regards.

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <gd6k3j$515$1@fred.mathworks.com>...
> "Johan " <robert.mchugh@ips.invensys.com> wrote in message <gd6dfl$be8$1@fred.mathworks.com>...
>
> > Is there some vectorized scheme or existing function that I should look into?
>
> help diff
>
> Bruno

"Johan " <robert.mchugh@ips.invensys.com> wrote in message <gd7lnq$h82$1@fred.mathworks.com>...

>
> Are there existing tools to smooth this out?
>
> Regards.
>

You might fit (avoid interpolate) the data with a model, then take a derivative of a fitted model. If you don't know the model, then using a universal cubic spline for model could better way than diff.

Bruno

"Johan " <robert.mchugh@ips.invensys.com> wrote in message <gd6dfl$be8$1@fred.mathworks.com>...
> Am looking for recommendations about how to create a vector of derivatives given a vector of values.
>
> In this case, I have an array with temperatures at different times (evenly spaced). I want to create an array of temperature derivatives. (I anticipate losing a few points at the ends.)
>
> Could write a loop, I suppose, dT/dt_i = (T_i+1 - T_i)/dt
> But there are 20,000 points.
>
> Is there some vectorized scheme or existing function that I should look into?
>
> Thanks.

Yes. Just look for a Savitsky-Golay tool. There
are many of them on the file exchange. Here
are a couple of them.

http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=16997&objectType=FILE

http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=4038&objectType=file

John