MathGroup Archive 1998

[Date Index] [Thread Index] [Author Index]

Search the Archive

High precision expressions and functions

The recent thread "High precision numbers and Plot", and the article
"Bending Plot to Your Needs" by Stan Wagon in Education and Research
Volume 7 No 2, Spring 1998 stimulated some experiments which resulted
in the following code

PrecisionExpression::usage = "PrecisionExpression[expr, {x1,x2..}, p]
for symbols x1,x2  and positive integer p gives an expression PEexpr
which, with x1=r1, x2 = r2,.., evaluates with working precision p,
starting with, xi = If[InexactNumberQ[vi], SetPrecision[vi, p], vi].
PrecisionExpression[expr, x, p], for a symbol x evaluates like
PrecisionExpression[expr, {x}, p] "

PrecisionExpression[expr_, x_, p_] :=
  Function[x, #][##2] &[
    expr /. r_?InexactNumberQ :> SetPrecision[r, p],
              _?InexactNumberQ, SetPrecision[#, p],
              _?NumberQ, #
              ] &/@Flatten[{x}]

PrecisionFunction::usage = "PrecisionFunction[expr, {x1,x2..}, p] for
symbols x1,x2  and positive integer p gives a function FPexpr with
working precision p. PFexpr[v1,v2,..] becomes
PFexpr[If[InexactNumberQ[v1], SetPrecision[v1, p], vi] ,..].
PrecisionFunction[expr, x, p], for a symbol x evaluates like
PrecisionFunction[expr, {x}, p]. "

PrecisionFunction[expr_, x_, p_] :=
  Function[x, Function[x, #][##2]] &[
   expr /. r_?InexactNumberQ :> SetPrecision[r,p],
              _?InexactNumberQ, SetPrecision[#, p],
              _?NumberQ, #
              ] &/@Flatten[{x}]

PrecisionExpression is the more flexible, but usually gives a slightly
slower result.



f1 = Normal[Series[Cos[x], {x, 0, 200}]];

expr = PrecisionExpression[f1, {x}, 30];

Block[{x = 1.2}, Do[expr, {50}] // Timing]

    {6.26 Second, Null}

foo = PrecisionFunction[f1, {x}, 30];

Do[foo[1.2], {50}] // Timing

    {4.56 Second, Null}

(2) Applications

f2= Log[1. - Erf[x]] + x^2;

Plot[#[x], {x, 0, 8}] &[PrecisionFunction[f2, x, 30]]

Plot[Evaluate[PrecisionExpression[{f2, Sin[x]-1.5}, x, 30]], {x, 0, 8}]

Plot3D[Evaluate[PrecisionExpression[f2 Sin[y], {x,y}, 30]], {x, 0,

ContourPlot[Evaluate[PrecisionExpression[f2 Sin[y], {x,y}, 30]], {x, 0,

Note: compiling is not used (it would spoil the precision), but we might
take the precaution of setting Compiled -> False. The (brief)time used
for checking if compiling is possible would also be save.

------------------------------------------------------------- Allan
Training and Consulting
Leicester UK
hay at
voice: +44 (0)116 271 4198
fax: +44(0)116 271 8642

  • Prev by Date: groebner Basis
  • Next by Date: Removing packages
  • Previous by thread: groebner Basis
  • Next by thread: Removing packages