Math.E Champ
Définition
Important
Certaines informations portent sur la préversion du produit qui est susceptible d’être en grande partie modifiée avant sa publication. Microsoft exclut toute garantie, expresse ou implicite, concernant les informations fournies ici.
Représente la base logarithmique naturelle spécifiée par la constante e
.
public: double E = 2.7182818284590451;
public const double E = 2.7182818284590451;
val mutable E : double
Public Const E As Double = 2.7182818284590451
Valeur de champ
Value = 2.7182818284590451Exemples
L’exemple suivant compare E la valeur calculée à partir d’une série d’alimentation.
// Example for the Math::E field.
using namespace System;
// Approximate E with a power series.
void CalcPowerSeries()
{
double factorial = 1.0;
double PS = 0.0;
// Stop iterating when the series converges,
// and prevent a runaway process.
for ( int n = 0; n < 999 && Math::Abs( Math::E - PS ) > 1.0E-15; n++ )
{
// Calculate a running factorial.
if ( n > 0 )
factorial *= (double)n;
// Calculate and display the power series.
PS += 1.0 / factorial;
Console::WriteLine( "PS({0:D2}) == {1:E16}, Math::E - PS({0:D2}) == {2:E16}", n, PS, Math::E - PS );
}
}
int main()
{
Console::WriteLine( "This example of Math::E == {0:E16}\n"
"generates the following output.\n", Math::E );
Console::WriteLine( "Define the power series PS(n) = Sum(k->0,n)[1/k!]" );
Console::WriteLine( " (limit n->infinity)PS(n) == e" );
Console::WriteLine( "Display PS(n) and Math::E - PS(n), "
"and stop when delta < 1.0E-15\n" );
CalcPowerSeries();
}
/*
This example of Math::E == 2.7182818284590451E+000
generates the following output.
Define the power series PS(n) = Sum(k->0,n)[1/k!]
(limit n->infinity)PS(n) == e
Display PS(n) and Math::E - PS(n), and stop when delta < 1.0E-15
PS(00) == 1.0000000000000000E+000, Math::E - PS(00) == 1.7182818284590451E+000
PS(01) == 2.0000000000000000E+000, Math::E - PS(01) == 7.1828182845904509E-001
PS(02) == 2.5000000000000000E+000, Math::E - PS(02) == 2.1828182845904509E-001
PS(03) == 2.6666666666666665E+000, Math::E - PS(03) == 5.1615161792378572E-002
PS(04) == 2.7083333333333330E+000, Math::E - PS(04) == 9.9484951257120535E-003
PS(05) == 2.7166666666666663E+000, Math::E - PS(05) == 1.6151617923787498E-003
PS(06) == 2.7180555555555554E+000, Math::E - PS(06) == 2.2627290348964380E-004
PS(07) == 2.7182539682539684E+000, Math::E - PS(07) == 2.7860205076724043E-005
PS(08) == 2.7182787698412700E+000, Math::E - PS(08) == 3.0586177750535626E-006
PS(09) == 2.7182815255731922E+000, Math::E - PS(09) == 3.0288585284310443E-007
PS(10) == 2.7182818011463845E+000, Math::E - PS(10) == 2.7312660577649694E-008
PS(11) == 2.7182818261984929E+000, Math::E - PS(11) == 2.2605521898810821E-009
PS(12) == 2.7182818282861687E+000, Math::E - PS(12) == 1.7287637987806193E-010
PS(13) == 2.7182818284467594E+000, Math::E - PS(13) == 1.2285727990501982E-011
PS(14) == 2.7182818284582302E+000, Math::E - PS(14) == 8.1490370007486490E-013
PS(15) == 2.7182818284589949E+000, Math::E - PS(15) == 5.0182080713057076E-014
PS(16) == 2.7182818284590429E+000, Math::E - PS(16) == 2.2204460492503131E-015
PS(17) == 2.7182818284590455E+000, Math::E - PS(17) == -4.4408920985006262E-016
*/
// Example for the Math.E field.
using System;
class EField
{
public static void Main()
{
Console.WriteLine(
"This example of Math.E == {0:E16}\n" +
"generates the following output.\n",
Math.E );
Console.WriteLine(
"Define the power series PS(n) = Sum(k->0,n)[1/k!]" );
Console.WriteLine( " (limit n->infinity)PS(n) == e" );
Console.WriteLine(
"Display PS(n) and Math.E - PS(n), " +
"and stop when delta < 1.0E-15\n" );
CalcPowerSeries();
}
// Approximate E with a power series.
static void CalcPowerSeries()
{
double factorial = 1.0;
double PS = 0.0;
// Stop iterating when the series converges,
// and prevent a runaway process.
for( int n = 0; n < 999 && Math.Abs( Math.E - PS ) > 1.0E-15; n++ )
{
// Calculate a running factorial.
if( n > 0 )
factorial *= (double)n;
// Calculate and display the power series.
PS += 1.0 / factorial;
Console.WriteLine(
"PS({0:D2}) == {1:E16}, Math.E - PS({0:D2}) == {2:E16}",
n, PS, Math.E - PS );
}
}
}
/*
This example of Math.E == 2.7182818284590451E+000
generates the following output.
Define the power series PS(n) = Sum(k->0,n)[1/k!]
(limit n->infinity)PS(n) == e
Display PS(n) and Math.E - PS(n), and stop when delta < 1.0E-15
PS(00) == 1.0000000000000000E+000, Math.E - PS(00) == 1.7182818284590451E+000
PS(01) == 2.0000000000000000E+000, Math.E - PS(01) == 7.1828182845904509E-001
PS(02) == 2.5000000000000000E+000, Math.E - PS(02) == 2.1828182845904509E-001
PS(03) == 2.6666666666666665E+000, Math.E - PS(03) == 5.1615161792378572E-002
PS(04) == 2.7083333333333330E+000, Math.E - PS(04) == 9.9484951257120535E-003
PS(05) == 2.7166666666666663E+000, Math.E - PS(05) == 1.6151617923787498E-003
PS(06) == 2.7180555555555554E+000, Math.E - PS(06) == 2.2627290348964380E-004
PS(07) == 2.7182539682539684E+000, Math.E - PS(07) == 2.7860205076724043E-005
PS(08) == 2.7182787698412700E+000, Math.E - PS(08) == 3.0586177750535626E-006
PS(09) == 2.7182815255731922E+000, Math.E - PS(09) == 3.0288585284310443E-007
PS(10) == 2.7182818011463845E+000, Math.E - PS(10) == 2.7312660577649694E-008
PS(11) == 2.7182818261984929E+000, Math.E - PS(11) == 2.2605521898810821E-009
PS(12) == 2.7182818282861687E+000, Math.E - PS(12) == 1.7287637987806193E-010
PS(13) == 2.7182818284467594E+000, Math.E - PS(13) == 1.2285727990501982E-011
PS(14) == 2.7182818284582302E+000, Math.E - PS(14) == 8.1490370007486490E-013
PS(15) == 2.7182818284589949E+000, Math.E - PS(15) == 5.0182080713057076E-014
PS(16) == 2.7182818284590429E+000, Math.E - PS(16) == 2.2204460492503131E-015
PS(17) == 2.7182818284590455E+000, Math.E - PS(17) == -4.4408920985006262E-016
*/
// Example for the Math.E field.
open System
// Approximate E with a power series.
let calcPowerSeries () =
let mutable factorial = 1.
let mutable PS = 0.
let mutable n = 0
// Stop iterating when the series converges,
// and prevent a runaway process.
while n < 999 && abs (Math.E - PS) > 1.0E-15 do
// Calculate a running factorial.
if n > 0 then
factorial <- factorial * double n
// Calculate and display the power series.
PS <- PS + 1. / factorial
printfn $"PS({n:D2}) = {PS:E16}, Math.E - PS({n:D2}) = {Math.E - PS:E16}"
n <- n + 1
printfn $"This example of Math.E = {Math.E:E16}\ngenerates the following output.\n"
printfn "Define the power series PS(n) = Sum(k->0,n)[1/k!]"
printfn " (limit n->infinity)PS(n) = e"
printfn "Display PS(n) and Math.E - PS(n), and stop when delta < 1.0E-15\n"
calcPowerSeries ()
// This example of Math.E = 2.7182818284590451E+000
// generates the following output.
//
// Define the power series PS(n) = Sum(k->0,n)[1/k!]
// (limit n->infinity)PS(n) = e
// Display PS(n) and Math.E - PS(n), and stop when delta < 1.0E-15
//
// PS(00) = 1.0000000000000000E+000, Math.E - PS(00) = 1.7182818284590451E+000
// PS(01) = 2.0000000000000000E+000, Math.E - PS(01) = 7.1828182845904509E-001
// PS(02) = 2.5000000000000000E+000, Math.E - PS(02) = 2.1828182845904509E-001
// PS(03) = 2.6666666666666665E+000, Math.E - PS(03) = 5.1615161792378572E-002
// PS(04) = 2.7083333333333330E+000, Math.E - PS(04) = 9.9484951257120535E-003
// PS(05) = 2.7166666666666663E+000, Math.E - PS(05) = 1.6151617923787498E-003
// PS(06) = 2.7180555555555554E+000, Math.E - PS(06) = 2.2627290348964380E-004
// PS(07) = 2.7182539682539684E+000, Math.E - PS(07) = 2.7860205076724043E-005
// PS(08) = 2.7182787698412700E+000, Math.E - PS(08) = 3.0586177750535626E-006
// PS(09) = 2.7182815255731922E+000, Math.E - PS(09) = 3.0288585284310443E-007
// PS(10) = 2.7182818011463845E+000, Math.E - PS(10) = 2.7312660577649694E-008
// PS(11) = 2.7182818261984929E+000, Math.E - PS(11) = 2.2605521898810821E-009
// PS(12) = 2.7182818282861687E+000, Math.E - PS(12) = 1.7287637987806193E-010
// PS(13) = 2.7182818284467594E+000, Math.E - PS(13) = 1.2285727990501982E-011
// PS(14) = 2.7182818284582302E+000, Math.E - PS(14) = 8.1490370007486490E-013
// PS(15) = 2.7182818284589949E+000, Math.E - PS(15) = 5.0182080713057076E-014
// PS(16) = 2.7182818284590429E+000, Math.E - PS(16) = 2.2204460492503131E-015
// PS(17) = 2.7182818284590455E+000, Math.E - PS(17) = -4.4408920985006262E-016
' Example for the Math.E field.
Module EField
Sub Main()
Console.WriteLine( _
"This example of Math.E = {0:E16}" & vbCrLf & _
"generates the following output." & vbCrLf, _
Math.E )
Console.WriteLine( _
"Define the power series PS(n) = Sum(k->0,n)[1/k!]" )
Console.WriteLine( " (limit n->infinity)PS(n) = e" )
Console.WriteLine( _
"Display PS(n) and Math.E - PS(n), " & _
"and stop when delta < 1.0E-15" & vbCrLf )
CalcPowerSeries()
End Sub
' Approximate E with a power series.
Sub CalcPowerSeries()
Dim factorial As Double = 1.0
Dim PS As Double = 0.0
' Stop iterating when the series converges,
' and prevent a runaway process.
Dim n As Integer
For n = 0 To 999
' Calculate a running factorial.
If n > 0 Then
factorial *= System.Convert.ToDouble(n)
End If
' Calculate and display the power series.
PS += 1.0 / factorial
Console.WriteLine( _
"PS({0:D2}) = {1:E16}, Math.E - PS({0:D2}) = {2:E16}", _
n, PS, Math.E - PS )
' Exit when the series converges.
If Math.Abs( Math.E - PS ) < 1.0E-15 Then
Exit For
End If
Next n
End Sub
End Module 'EField
' This example of Math.E = 2.7182818284590451E+000
' generates the following output.
'
' Define the power series PS(n) = Sum(k->0,n)[1/k!]
' (limit n->infinity)PS(n) = e
' Display PS(n) and Math.E - PS(n), and stop when delta < 1.0E-15
'
' PS(00) = 1.0000000000000000E+000, Math.E - PS(00) = 1.7182818284590451E+000
' PS(01) = 2.0000000000000000E+000, Math.E - PS(01) = 7.1828182845904509E-001
' PS(02) = 2.5000000000000000E+000, Math.E - PS(02) = 2.1828182845904509E-001
' PS(03) = 2.6666666666666665E+000, Math.E - PS(03) = 5.1615161792378572E-002
' PS(04) = 2.7083333333333330E+000, Math.E - PS(04) = 9.9484951257120535E-003
' PS(05) = 2.7166666666666663E+000, Math.E - PS(05) = 1.6151617923787498E-003
' PS(06) = 2.7180555555555554E+000, Math.E - PS(06) = 2.2627290348964380E-004
' PS(07) = 2.7182539682539684E+000, Math.E - PS(07) = 2.7860205076724043E-005
' PS(08) = 2.7182787698412700E+000, Math.E - PS(08) = 3.0586177750535626E-006
' PS(09) = 2.7182815255731922E+000, Math.E - PS(09) = 3.0288585284310443E-007
' PS(10) = 2.7182818011463845E+000, Math.E - PS(10) = 2.7312660577649694E-008
' PS(11) = 2.7182818261984929E+000, Math.E - PS(11) = 2.2605521898810821E-009
' PS(12) = 2.7182818282861687E+000, Math.E - PS(12) = 1.7287637987806193E-010
' PS(13) = 2.7182818284467594E+000, Math.E - PS(13) = 1.2285727990501982E-011
' PS(14) = 2.7182818284582302E+000, Math.E - PS(14) = 8.1490370007486490E-013
' PS(15) = 2.7182818284589949E+000, Math.E - PS(15) = 5.0182080713057076E-014
' PS(16) = 2.7182818284590429E+000, Math.E - PS(16) = 2.2204460492503131E-015
' PS(17) = 2.7182818284590455E+000, Math.E - PS(17) = -4.4408920985006262E-016
Remarques
La valeur de ce champ est 2,7182818284590451.
S’applique à
Collaborer avec nous sur GitHub
La source de ce contenu se trouve sur GitHub, où vous pouvez également créer et examiner les problèmes et les demandes de tirage (pull requests). Pour plus d’informations, consultez notre guide du contributeur.