Higher precision in MATLAB App Designer

I am developing a GUI with App Designer and the calculations need precision upto 16 decimal places. Under normal circumstances "format long" can be used to increase the precision. How can high precision be obtained in the App environment?

Answer 1

[Steven Lord]

The format function does not "increase the precision" of calculations. All it does is change how the values are displayed in the Command Window. If you want control over how the numbers are displayed in App Designer take a look at the sprintf function.

[Tanu Singla]

Thank you, Steven. I was under the impression that format function actually increases the precision. What should I do to increase the precision for the calculation in App Designer and not just display the numbers.

[Steven Lord]

Can you say more about the problem you're trying to solve and why you feel that you need to use greater than double precision to solve it?

[Walter Roberson]

App Designer code can use the symbolic toolbox same as regular code. (But as per usual, anything built with the symbolic toolbox cannot be compiled.)

[Tanu Singla]

Steven, I am developing a GUI to create Mandelbrot sets which requires high level zoom in to the xy plot. With double precision, results cannot be obtained after a certain zoom.

[Walter Roberson]

Note that Mandelbrot may act differently with symbolic computation. Symbolic computation is modeled as if it were carried out in decimal, but if you search a while you can find references indicating that the computation is done in groups of 9 decimal digits (10^10 is just under 2^30).

But if you push the tests hard enough that doesn't quite work out and the implication would seem to be that rationals are stored in groups of 30 bits rather than in decimal. Design documents that talk about this are... scarce.

Symbolic floating point computation is carried out with 5 guard digits beyond the digits() setting, and appears to be calculated in decimal... maybe.

Answer 2

[Steven Lord]

How far do you expect your users to want to zoom in? What do you expect typical limits of the region of interest to be? If it falls within the limits of quadruple precision you may be able to calculate the results using Cleve's prototype or a MEX-file doing quad precision calculations in C or C++, but I don't know if you'll be able to plot it. My guess is that figures, axes, etc. won't be very happy trying to display numbers on the order of say 1e-4000.

[Walter Roberson]

In my experience, plots are done at single precision internally (though I no longer recall the context for that.)

[Tanu Singla]

I think quadruple precision should do just fine in my case.

Plots can be obtained at single precision but the way I am developing the GUI, it obtaines the x and y interval from the zoom in region of the plot only. If axes canot handle higher precision,I might have to change my algorithm.