2
$\begingroup$

I have the following function:

L[n_, t_] := Abs[2/Pi*(2 n - 1)!!^2/((2 n)! (2 n - 1)^2)  *   
HypergeometricPFQ[{3/2, 1/2 - n, 1/2 - n}, {3/2 - n, 3/2 - n}, Exp[-2 I*t]]]^2

If I now plot this function, let's say: 

Plot[L[n, t], {t, 0, Pi/2}, PerformanceGoal -> "Speed"]

Then, very "smooth" results are yielded until n=16. For n>16, the plot becomes really oscillatory, which seems weird to me.

I wonder whether this is a numeric problem and how to find out whether it is? I evaluated some values with high precision and they seem to agree with the plot... This is what I tested, for example:

 Lo[N[17, 200], N[0.273, 100]]=0.
Improve this question
$\endgroup$
  • 4
    $\begingroup$ You're observing numerical instabilities in fixed-precision arithmetic. Take advantage of Mathematica's arbitrary precision capabilities by adding the option WorkingPrecision->20 to the Plot. Raise the setting as needed to address any observed instabilities. $\endgroup$ – QuantumDot Jul 20 '17 at 13:35
3
$\begingroup$

Apperantely the machine precision can be limiting, see the help function for HypergeometricPFQ.

Machine-precision input may be insufficient to get a correct answer:

In[1]:= HypergeometricPFQ[{10, 10}, {50}, 2.]

Out[1]= -12642.7 + 12642.7 I

With exact input, the answer is correct:

In[2]:= N[HypergeometricPFQ[{10, 10}, {50}, 2], 20]

Out[2]= -1705.7573316355444417 - 356.7499860560238175 I

If we plot it with exact inputs, i.e.

ListPlot[Table[{t, L[17, t]}, {t, 0, π/2, π/100}]]

we get the right result:

discrete plot

We can also increase WorkingPrecision like QuantumDot suggests:

Plot[L[17, t], {t, 0, π/2}, WorkingPrecision -> 30]

high-precision plot

Improve this answer
$\endgroup$
  • $\begingroup$ The diagnosis is essentially right. See my comment to the OP to effect a cure to Plot. $\endgroup$ – QuantumDot Jul 20 '17 at 13:37
4
$\begingroup$

One approach is to use the adaptive precision control built into N. Three digits is usually enough for plotting. Start by setting the precision of the input to Infinity, which converts it to the exact fraction represented by the binary floating-point number:

Block[{f},
  f[t_?NumericQ] := With[{tt = SetPrecision[t, Infinity]}, N[L[17, tt], 3]];
  Plot[f[t], {t, 0, Pi/2}]
  ] // AbsoluteTiming

Mathematica graphics

It's faster than a set arbitrary precision:

Plot[L[17, t], {t, 0, Pi/2}, 
  WorkingPrecision -> 20] // AbsoluteTiming

Mathematica graphics

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.