phobos/std/complex.d
2012-05-11 19:08:12 +02:00

713 lines
17 KiB
D

// Written in the D programming language.
/** This module contains the $(LREF Complex) type, which is used to represent
_complex numbers, along with related mathematical operations and functions.
$(LREF Complex) will eventually $(LINK2 ../deprecate.html, replace)
the built-in types $(D cfloat), $(D cdouble), $(D creal), $(D ifloat),
$(D idouble), and $(D ireal).
Authors: Lars Tandle Kyllingstad, Don Clugston
Copyright: Copyright (c) 2010, Lars T. Kyllingstad.
License: $(WEB boost.org/LICENSE_1_0.txt, Boost License 1.0)
Source: $(PHOBOSSRC std/_complex.d)
*/
module std.complex;
import std.format;
import std.math;
import std.numeric;
import std.traits;
/** Helper function that returns a _complex number with the specified
real and imaginary parts.
If neither $(D re) nor $(D im) are floating-point numbers, this
function returns a $(D Complex!double). Otherwise, the return type
is deduced using $(D std.traits.CommonType!(R, I)).
Examples:
---
auto c = complex(2.0);
static assert (is(typeof(c) == Complex!double));
assert (c.re == 2.0);
assert (c.im == 0.0);
auto w = complex(2);
static assert (is(typeof(w) == Complex!double));
assert (w == c);
auto z = complex(1, 3.14L);
static assert (is(typeof(z) == Complex!real));
assert (z.re == 1.0L);
assert (z.im == 3.14L);
---
*/
auto complex(T)(T re) @safe pure nothrow if (is(T : double))
{
static if (isFloatingPoint!T)
return Complex!T(re, 0);
else
return Complex!double(re, 0);
}
/// ditto
auto complex(R, I)(R re, I im) @safe pure nothrow
if (is(R : double) && is(I : double))
{
static if (isFloatingPoint!R || isFloatingPoint!I)
return Complex!(CommonType!(R, I))(re, im);
else
return Complex!double(re, im);
}
unittest
{
auto a = complex(1.0);
static assert (is(typeof(a) == Complex!double));
assert (a.re == 1.0);
assert (a.im == 0.0);
auto b = complex(2.0L);
static assert (is(typeof(b) == Complex!real));
assert (b.re == 2.0L);
assert (b.im == 0.0L);
auto c = complex(1.0, 2.0);
static assert (is(typeof(c) == Complex!double));
assert (c.re == 1.0);
assert (c.im == 2.0);
auto d = complex(3.0, 4.0L);
static assert (is(typeof(d) == Complex!real));
assert (d.re == 3.0);
assert (d.im == 4.0L);
auto e = complex(1);
static assert (is(typeof(e) == Complex!double));
assert (e.re == 1);
assert (e.im == 0);
auto f = complex(1L, 2);
static assert (is(typeof(f) == Complex!double));
assert (f.re == 1L);
assert (f.im == 2);
auto g = complex(3, 4.0L);
static assert (is(typeof(g) == Complex!real));
assert (g.re == 3);
assert (g.im == 4.0L);
}
/** A complex number parametrised by a type $(D T), which must be either
$(D float), $(D double) or $(D real).
*/
struct Complex(T) if (isFloatingPoint!T)
{
/** The real part of the number. */
T re;
/** The imaginary part of the number. */
T im;
/** Converts the complex number to a string representation.
If a $(D sink) delegate is specified, the string is passed to it
and this function returns $(D null). Otherwise, this function
returns the string representation directly.
The output format is controlled via $(D formatSpec), which should consist
of a single POSIX format specifier, including the percent (%) character.
Note that complex numbers are floating point numbers, so the only
valid format characters are 'e', 'f', 'g', 'a', and 's', where 's'
gives the default behaviour. Positional parameters are not valid
in this context.
See the $(LINK2 std_format.html, std.format documentation) for
more information.
*/
string toString(scope void delegate(const(char)[]) sink = null,
string formatSpec = "%s")
const
{
if (sink == null)
{
char[] buf;
buf.reserve(100);
toString((const(char)[] s) { buf ~= s; }, formatSpec);
return cast(string) buf;
}
formattedWrite(sink, formatSpec, re);
if (signbit(im) == 0) sink("+");
formattedWrite(sink, formatSpec, im);
sink("i");
return null;
}
@safe pure nothrow:
// ASSIGNMENT OPERATORS
// this = complex
ref Complex opAssign(R : T)(Complex!R z)
{
re = z.re;
im = z.im;
return this;
}
// this = numeric
ref Complex opAssign(R : T)(R r)
{
re = r;
im = 0;
return this;
}
// COMPARISON OPERATORS
// this == complex
bool opEquals(R : T)(Complex!R z) const
{
return re == z.re && im == z.im;
}
// this == numeric
bool opEquals(R : T)(R r) const
{
return re == r && im == 0;
}
// UNARY OPERATORS
// +complex
Complex opUnary(string op)() const
if (op == "+")
{
return this;
}
// -complex
Complex opUnary(string op)() const
if (op == "-")
{
return Complex(-re, -im);
}
// BINARY OPERATORS
// complex op complex
Complex!(CommonType!(T,R)) opBinary(string op, R)(Complex!R z) const
{
alias typeof(return) C;
auto w = C(this.re, this.im);
return w.opOpAssign!(op)(z);
}
// complex op numeric
Complex!(CommonType!(T,R)) opBinary(string op, R)(R r) const
if (isNumeric!R)
{
alias typeof(return) C;
auto w = C(this.re, this.im);
return w.opOpAssign!(op)(r);
}
// numeric + complex, numeric * complex
Complex!(CommonType!(T, R)) opBinaryRight(string op, R)(R r) const
if ((op == "+" || op == "*") && (isNumeric!R))
{
return opBinary!(op)(r);
}
// numeric - complex
Complex!(CommonType!(T, R)) opBinaryRight(string op, R)(R r) const
if (op == "-" && isNumeric!R)
{
return Complex(r - re, -im);
}
// numeric / complex
Complex!(CommonType!(T, R)) opBinaryRight(string op, R)(R r) const
if (op == "/" && isNumeric!R)
{
typeof(return) w;
alias FPTemporary!(typeof(w.re)) Tmp;
if (fabs(re) < fabs(im))
{
Tmp ratio = re/im;
Tmp rdivd = r/(re*ratio + im);
w.re = rdivd*ratio;
w.im = -rdivd;
}
else
{
Tmp ratio = im/re;
Tmp rdivd = r/(re + im*ratio);
w.re = rdivd;
w.im = -rdivd*ratio;
}
return w;
}
// OP-ASSIGN OPERATORS
// complex += complex, complex -= complex
ref Complex opOpAssign(string op, C)(C z)
if ((op == "+" || op == "-") && is(C R == Complex!R))
{
mixin ("re "~op~"= z.re;");
mixin ("im "~op~"= z.im;");
return this;
}
// complex *= complex
ref Complex opOpAssign(string op, C)(C z)
if (op == "*" && is(C R == Complex!R))
{
auto temp = re*z.re - im*z.im;
im = im*z.re + re*z.im;
re = temp;
return this;
}
// complex /= complex
ref Complex opOpAssign(string op, C)(C z)
if (op == "/" && is(C R == Complex!R))
{
if (fabs(z.re) < fabs(z.im))
{
FPTemporary!T ratio = z.re/z.im;
FPTemporary!T denom = z.re*ratio + z.im;
auto temp = (re*ratio + im)/denom;
im = (im*ratio - re)/denom;
re = temp;
}
else
{
FPTemporary!T ratio = z.im/z.re;
FPTemporary!T denom = z.re + z.im*ratio;
auto temp = (re + im*ratio)/denom;
im = (im - re*ratio)/denom;
re = temp;
}
return this;
}
// complex ^^= complex
ref Complex opOpAssign(string op, C)(C z)
if (op == "^^" && is(C R == Complex!R))
{
FPTemporary!T r = abs(this);
FPTemporary!T t = arg(this);
FPTemporary!T ab = r^^z.re * exp(-t*z.im);
FPTemporary!T ar = t*z.re + log(r)*z.im;
re = ab*std.math.cos(ar);
im = ab*std.math.sin(ar);
return this;
}
// complex += numeric, complex -= numeric
ref Complex opOpAssign(string op, U : T)(U a)
if (op == "+" || op == "-")
{
mixin ("re "~op~"= a;");
return this;
}
// complex *= numeric, complex /= numeric
ref Complex opOpAssign(string op, U : T)(U a)
if (op == "*" || op == "/")
{
mixin ("re "~op~"= a;");
mixin ("im "~op~"= a;");
return this;
}
// complex ^^= real
ref Complex opOpAssign(string op, R)(R r)
if (op == "^^" && isFloatingPoint!R)
{
FPTemporary!T ab = abs(this)^^r;
FPTemporary!T ar = arg(this)*r;
re = ab*std.math.cos(ar);
im = ab*std.math.sin(ar);
return this;
}
// complex ^^= int
ref Complex opOpAssign(string op, U)(U i)
if (op == "^^" && isIntegral!U)
{
switch (i)
{
case 0:
re = 1.0;
im = 0.0;
break;
case 1:
// identity; do nothing
break;
case 2:
this *= this;
break;
case 3:
auto z = this;
this *= z;
this *= z;
break;
default:
this ^^= cast(real) i;
}
return this;
}
}
unittest
{
enum EPS = double.epsilon;
auto c1 = complex(1.0, 1.0);
// Check unary operations.
auto c2 = Complex!double(0.5, 2.0);
assert (c2 == +c2);
assert ((-c2).re == -(c2.re));
assert ((-c2).im == -(c2.im));
assert (c2 == -(-c2));
// Check complex-complex operations.
auto cpc = c1 + c2;
assert (cpc.re == c1.re + c2.re);
assert (cpc.im == c1.im + c2.im);
auto cmc = c1 - c2;
assert (cmc.re == c1.re - c2.re);
assert (cmc.im == c1.im - c2.im);
auto ctc = c1 * c2;
assert (approxEqual(abs(ctc), abs(c1)*abs(c2), EPS));
assert (approxEqual(arg(ctc), arg(c1)+arg(c2), EPS));
auto cdc = c1 / c2;
assert (approxEqual(abs(cdc), abs(c1)/abs(c2), EPS));
assert (approxEqual(arg(cdc), arg(c1)-arg(c2), EPS));
auto cec = c1^^c2;
assert (approxEqual(cec.re, 0.11524131979943839881, EPS));
assert (approxEqual(cec.im, 0.21870790452746026696, EPS));
// Check complex-real operations.
double a = 123.456;
auto cpr = c1 + a;
assert (cpr.re == c1.re + a);
assert (cpr.im == c1.im);
auto cmr = c1 - a;
assert (cmr.re == c1.re - a);
assert (cmr.im == c1.im);
auto ctr = c1 * a;
assert (ctr.re == c1.re*a);
assert (ctr.im == c1.im*a);
auto cdr = c1 / a;
assert (approxEqual(abs(cdr), abs(c1)/a, EPS));
assert (approxEqual(arg(cdr), arg(c1), EPS));
auto rpc = a + c1;
assert (rpc == cpr);
auto rmc = a - c1;
assert (rmc.re == a-c1.re);
assert (rmc.im == -c1.im);
auto rtc = a * c1;
assert (rtc == ctr);
auto rdc = a / c1;
assert (approxEqual(abs(rdc), a/abs(c1), EPS));
assert (approxEqual(arg(rdc), -arg(c1), EPS));
auto cer = c1^^3.0;
assert (approxEqual(abs(cer), abs(c1)^^3, EPS));
assert (approxEqual(arg(cer), arg(c1)*3, EPS));
// Check Complex-int operations.
foreach (i; 0..6)
{
auto cei = c1^^i;
assert (approxEqual(abs(cei), abs(c1)^^i, EPS));
// Use cos() here to deal with arguments that go outside
// the (-pi,pi] interval (only an issue for i>3).
assert (approxEqual(std.math.cos(arg(cei)), std.math.cos(arg(c1)*i), EPS));
}
// Check operations between different complex types.
auto cf = Complex!float(1.0, 1.0);
auto cr = Complex!real(1.0, 1.0);
auto c1pcf = c1 + cf;
auto c1pcr = c1 + cr;
static assert (is(typeof(c1pcf) == Complex!double));
static assert (is(typeof(c1pcr) == Complex!real));
assert (c1pcf.re == c1pcr.re);
assert (c1pcf.im == c1pcr.im);
}
unittest
{
// Assignments and comparisons
Complex!double z;
z = 1;
assert (z == 1);
assert (z.re == 1.0 && z.im == 0.0);
z = 2.0;
assert (z == 2.0);
assert (z.re == 2.0 && z.im == 0.0);
z = 1.0L;
assert (z == 1.0L);
assert (z.re == 1.0 && z.im == 0.0);
auto w = Complex!real(1.0, 1.0);
z = w;
assert (z == w);
assert (z.re == 1.0 && z.im == 1.0);
auto c = Complex!float(2.0, 2.0);
z = c;
assert (z == c);
assert (z.re == 2.0 && z.im == 2.0);
}
unittest
{
// Convert to string.
// Using default format specifier
auto z1 = Complex!real(0.123456789, 0.123456789);
char[] s1;
z1.toString((const(char)[] c) { s1 ~= c; });
assert (s1 == "0.123457+0.123457i");
assert (s1 == z1.toString());
// Using custom format specifier
auto z2 = conj(z1);
char[] s2;
z2.toString((const(char)[] c) { s2 ~= c; }, "%.8e");
assert (s2 == "1.23456789e-01-1.23456789e-01i");
assert (s2 == z2.toString(null, "%.8e"));
}
/* Makes Complex!(Complex!T) fold to Complex!T.
The rationale for this is that just like the real line is a
subspace of the complex plane, the complex plane is a subspace
of itself. Example of usage:
---
Complex!T addI(T)(T x)
{
return x + Complex!T(0.0, 1.0);
}
---
The above will work if T is both real and complex.
*/
template Complex(T) if (is(T R == Complex!R))
{
alias T Complex;
}
unittest
{
static assert (is(Complex!(Complex!real) == Complex!real));
Complex!T addI(T)(T x)
{
return x + Complex!T(0.0, 1.0);
}
auto z1 = addI(1.0);
assert (z1.re == 1.0 && z1.im == 1.0);
enum one = Complex!double(1.0, 0.0);
auto z2 = addI(one);
assert (z1 == z2);
}
/** Calculates the absolute value (or modulus) of a complex number. */
T abs(T)(Complex!T z) @safe pure nothrow
{
return hypot(z.re, z.im);
}
unittest
{
assert (abs(complex(1.0)) == 1.0);
assert (abs(complex(0.0, 1.0)) == 1.0);
assert (abs(complex(1.0L, -2.0L)) == std.math.sqrt(5.0L));
}
/** Calculates the argument (or phase) of a complex number. */
T arg(T)(Complex!T z) @safe pure nothrow
{
return atan2(z.im, z.re);
}
unittest
{
assert (arg(complex(1.0)) == 0.0);
assert (arg(complex(0.0L, 1.0L)) == PI_2);
assert (arg(complex(1.0L, 1.0L)) == PI_4);
}
/** Returns the complex conjugate of a complex number. */
Complex!T conj(T)(Complex!T z) @safe pure nothrow
{
return Complex!T(z.re, -z.im);
}
unittest
{
assert (conj(complex(1.0)) == complex(1.0));
assert (conj(complex(1.0, 2.0)) == complex(1.0, -2.0));
}
/** Constructs a complex number given its absolute value and argument. */
Complex!(CommonType!(T, U)) fromPolar(T, U)(T modulus, U argument)
@safe pure nothrow
{
return Complex!(CommonType!(T,U))
(modulus*std.math.cos(argument), modulus*std.math.sin(argument));
}
unittest
{
auto z = fromPolar(std.math.sqrt(2.0), PI_4);
assert (approxEqual(z.re, 1.0L, real.epsilon));
assert (approxEqual(z.im, 1.0L, real.epsilon));
}
/** Trigonometric functions. */
Complex!T sin(T)(Complex!T z) @safe pure nothrow
{
auto cs = expi(z.re);
auto csh = coshisinh(z.im);
return typeof(return)(cs.im * csh.re, cs.re * csh.im);
}
unittest
{
assert(sin(complex(0.0)) == 0.0);
assert(sin(complex(2.0L, 0)) == std.math.sin(2.0L));
}
/// ditto
Complex!T cos(T)(Complex!T z) @safe pure nothrow
{
auto cs = expi(z.re);
auto csh = coshisinh(z.im);
return typeof(return)(cs.re * csh.re, - cs.im * csh.im);
}
unittest{
assert(cos(complex(0.0)) == 1.0);
assert(cos(complex(1.3L)) == std.math.cos(1.3L));
assert(cos(complex(0, 5.2L)) == cosh(5.2L));
}
/** Calculates cos(y) + i sin(y).
Note:
$(D expi) is included here for convenience and for easy migration of code
that uses $(XREF math,_expi). Unlike $(XREF math,_expi), which uses the
x87 $(I fsincos) instruction when possible, this function is no faster
than calculating cos(y) and sin(y) separately.
*/
Complex!real expi(real y) @trusted pure nothrow
{
return Complex!real(std.math.cos(y), std.math.sin(y));
}
unittest
{
assert(expi(1.3e5L) == complex(std.math.cos(1.3e5L), std.math.sin(1.3e5L)));
assert(expi(0.0L) == 1.0L);
auto z1 = expi(1.234);
auto z2 = std.math.expi(1.234);
assert(z1.re == z2.re && z1.im == z2.im);
}
/** Square root. */
Complex!T sqrt(T)(Complex!T z) @safe pure nothrow
{
typeof(return) c;
real x,y,w,r;
if (z == 0)
{
c = typeof(return)(0, 0);
}
else
{
real z_re = z.re;
real z_im = z.im;
x = fabs(z_re);
y = fabs(z_im);
if (x >= y)
{
r = y / x;
w = std.math.sqrt(x)
* std.math.sqrt(0.5 * (1 + std.math.sqrt(1 + r * r)));
}
else
{
r = x / y;
w = std.math.sqrt(y)
* std.math.sqrt(0.5 * (r + std.math.sqrt(1 + r * r)));
}
if (z_re >= 0)
{
c = typeof(return)(w, z_im / (w + w));
}
else
{
if (z_im < 0)
w = -w;
c = typeof(return)(z_im / (w + w), w);
}
}
return c;
}
unittest
{
assert (sqrt(complex(0.0)) == 0.0);
assert (sqrt(complex(1.0L, 0)) == std.math.sqrt(1.0L));
assert (sqrt(complex(-1.0L, 0)) == complex(0, 1.0L));
}