Plotting fits over significantly larger axis ranges

Hello,

I am encountering a strange behaviour of matlab, and couldn't find a way to force Matlab to do so.

Both graphs are fitting the excactly same fit, which i am polling from a structure.

The only difference is that for the graphs that look weird, I am plotting them over the axis [-2000 2000 -800 800], and then scale the figure back down to the actual xy-limits of the plotted data + margin on all sites [1 3 0.3 1.2]. The picture above is zoomed out a bit to see that the graphs are plotted outside the axis that I plotted the "finer" graphs over.

My current idea is that matlab obviously optimises the weird graphs shown above to look right on the axes you define:

Sadly, that would suggest that there is no way of optimise the plot over larger distances, and that I will have to adjust the factor for the first sketch down a lot, to get it to look properly again.

Or is there a method to force matlab to not simplify?

Thank you & stay healty.

Claudius Appel

[dpb]

The above makes no sense at all...MATLAB does not "simplify"; it plots what it's asked to plot over the ranges it's asked to plot over.

If you scale something with detail in it at the range of 1/5E5 --> ~0.001% of full scale, it's not surprising such detail will simply not be visible on the plot given there are only about 1000 pixels total on a typical PC monitor in the y-axis direction, the entire area of interest will be only about 1 pixel maximum. Since the typical figure and axis is only (say) roughly 1/4 of that, you've got everything well inside the one pixel boundary.

It is simply physically impossible to display such on the screen simultaneously on a linear scale.

[dpb]

Well, let's just see...

>> hAx=axes; hAx.Units="pixels"
hAx = 
  Axes with properties:

             XLim: [0 1.00]
             YLim: [0 1.00]
           XScale: 'linear'
           YScale: 'linear'
    GridLineStyle: '-'
         Position: [73.80 47.20 434.00 342.30]
            Units: 'pixels'

  Show all properties
>> 

OK, a default axes on the system here is 434 pixels high, not quite half the full screen height...what fraction of a pixel is the range of your expanded axis--

>> 2/4.5E5 * 434
ans =
    0.0019
>>

or, still in the noise of a thousandth of one pixel for the whole axis to be represented -- some 500 of them will fit in a single pixel, iow.

The scale of one pixel is roughly

>> round(4.5E5/434)
ans =
        1037
>> 

So any detail smaller than a difference of 1037 is indistinguishable.

[dpb]

BTW, you can always try

hAx=gca;                                % get current axes handle (if you didn't save it)
hAx.XScale='log'; hAx.YScale='log';     % Use log-log on large scale

will allocate same fraction of the axes space to each decade of data...(*)

The more normal way in such a case if the detail is significant is to use an inset axes like the city map on the state map in the road atlas.

(*) Of course, the (-)ive x-axis won't show on default log axis; there is a FEX submittal that mirrors the negative axis for the purpose but I don't recall the specific submission otomh.