aboutsummaryrefslogtreecommitdiff
path: root/v_windows/v/old/vlib/math/complex/complex_test.v
diff options
context:
space:
mode:
authorIndrajith K L2022-12-03 17:00:20 +0530
committerIndrajith K L2022-12-03 17:00:20 +0530
commitf5c4671bfbad96bf346bd7e9a21fc4317b4959df (patch)
tree2764fc62da58f2ba8da7ed341643fc359873142f /v_windows/v/old/vlib/math/complex/complex_test.v
downloadcli-tools-windows-master.tar.gz
cli-tools-windows-master.tar.bz2
cli-tools-windows-master.zip
Adds most of the toolsHEADmaster
Diffstat (limited to 'v_windows/v/old/vlib/math/complex/complex_test.v')
-rw-r--r--v_windows/v/old/vlib/math/complex/complex_test.v797
1 files changed, 797 insertions, 0 deletions
diff --git a/v_windows/v/old/vlib/math/complex/complex_test.v b/v_windows/v/old/vlib/math/complex/complex_test.v
new file mode 100644
index 0000000..ccd448e
--- /dev/null
+++ b/v_windows/v/old/vlib/math/complex/complex_test.v
@@ -0,0 +1,797 @@
+import math
+import math.complex as cmplx
+
+fn tst_res(str1 string, str2 string) bool {
+ if (math.abs(str1.f64() - str2.f64())) < 1e-5 {
+ return true
+ }
+ return false
+}
+
+fn test_complex_addition() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c1 := cmplx.complex(0, -10)
+ mut c2 := cmplx.complex(-40, 8)
+ mut result := c1 + c2
+ assert result.equals(cmplx.complex(-40, -2))
+ c1 = cmplx.complex(-71, 2)
+ c2 = cmplx.complex(88, -12)
+ result = c1 + c2
+ assert result.equals(cmplx.complex(17, -10))
+ c1 = cmplx.complex(0, -30)
+ c2 = cmplx.complex(52, -30)
+ result = c1 + c2
+ assert result.equals(cmplx.complex(52, -60))
+ c1 = cmplx.complex(12, -9)
+ c2 = cmplx.complex(32, -6)
+ result = c1 + c2
+ assert result.equals(cmplx.complex(44, -15))
+}
+
+fn test_complex_subtraction() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c1 := cmplx.complex(-8, 0)
+ mut c2 := cmplx.complex(6, 30)
+ mut result := c1 - c2
+ assert result.equals(cmplx.complex(-14, -30))
+ c1 = cmplx.complex(-19, 7)
+ c2 = cmplx.complex(29, 32)
+ result = c1 - c2
+ assert result.equals(cmplx.complex(-48, -25))
+ c1 = cmplx.complex(12, 0)
+ c2 = cmplx.complex(23, 13)
+ result = c1 - c2
+ assert result.equals(cmplx.complex(-11, -13))
+ c1 = cmplx.complex(-14, 3)
+ c2 = cmplx.complex(0, 14)
+ result = c1 - c2
+ assert result.equals(cmplx.complex(-14, -11))
+}
+
+fn test_complex_multiplication() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c1 := cmplx.complex(1, 2)
+ mut c2 := cmplx.complex(1, -4)
+ mut result := c1 * c2
+ assert result.equals(cmplx.complex(9, -2))
+ c1 = cmplx.complex(-4, -4)
+ c2 = cmplx.complex(-5, -3)
+ result = c1 * c2
+ assert result.equals(cmplx.complex(8, 32))
+ c1 = cmplx.complex(4, 4)
+ c2 = cmplx.complex(-2, -5)
+ result = c1 * c2
+ assert result.equals(cmplx.complex(12, -28))
+ c1 = cmplx.complex(2, -2)
+ c2 = cmplx.complex(4, -4)
+ result = c1 * c2
+ assert result.equals(cmplx.complex(0, -16))
+}
+
+fn test_complex_division() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c1 := cmplx.complex(-9, -6)
+ mut c2 := cmplx.complex(-3, -2)
+ mut result := c1 / c2
+ assert result.equals(cmplx.complex(3, 0))
+ c1 = cmplx.complex(-23, 11)
+ c2 = cmplx.complex(5, 1)
+ result = c1 / c2
+ assert result.equals(cmplx.complex(-4, 3))
+ c1 = cmplx.complex(8, -2)
+ c2 = cmplx.complex(-4, 1)
+ result = c1 / c2
+ assert result.equals(cmplx.complex(-2, 0))
+ c1 = cmplx.complex(11, 24)
+ c2 = cmplx.complex(-4, -1)
+ result = c1 / c2
+ assert result.equals(cmplx.complex(-4, -5))
+}
+
+fn test_complex_conjugate() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c1 := cmplx.complex(0, 8)
+ mut result := c1.conjugate()
+ assert result.equals(cmplx.complex(0, -8))
+ c1 = cmplx.complex(7, 3)
+ result = c1.conjugate()
+ assert result.equals(cmplx.complex(7, -3))
+ c1 = cmplx.complex(2, 2)
+ result = c1.conjugate()
+ assert result.equals(cmplx.complex(2, -2))
+ c1 = cmplx.complex(7, 0)
+ result = c1.conjugate()
+ assert result.equals(cmplx.complex(7, 0))
+}
+
+fn test_complex_equals() {
+ mut c1 := cmplx.complex(0, 8)
+ mut c2 := cmplx.complex(0, 8)
+ assert c1.equals(c2)
+ c1 = cmplx.complex(-3, 19)
+ c2 = cmplx.complex(-3, 19)
+ assert c1.equals(c2)
+}
+
+fn test_complex_abs() {
+ mut c1 := cmplx.complex(3, 4)
+ assert c1.abs() == 5
+ c1 = cmplx.complex(1, 2)
+ assert c1.abs() == math.sqrt(5)
+ assert c1.abs() == c1.conjugate().abs()
+ c1 = cmplx.complex(7, 0)
+ assert c1.abs() == 7
+}
+
+fn test_complex_angle() {
+ // Test is based on and verified from practice examples of Khan Academy
+ // https://www.khanacademy.org/math/precalculus/imaginary-and-complex-numbers
+ mut c := cmplx.complex(1, 0)
+ assert c.angle() * 180 / math.pi == 0
+ c = cmplx.complex(1, 1)
+ assert c.angle() * 180 / math.pi == 45
+ c = cmplx.complex(0, 1)
+ assert c.angle() * 180 / math.pi == 90
+ c = cmplx.complex(-1, 1)
+ assert c.angle() * 180 / math.pi == 135
+ c = cmplx.complex(-1, -1)
+ assert c.angle() * 180 / math.pi == -135
+ cc := c.conjugate()
+ assert cc.angle() + c.angle() == 0
+}
+
+fn test_complex_addinv() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-5, -7)
+ mut result := c1.addinv()
+ assert result.equals(c2)
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(3, -4)
+ result = c1.addinv()
+ assert result.equals(c2)
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(1, 2)
+ result = c1.addinv()
+ assert result.equals(c2)
+}
+
+fn test_complex_mulinv() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.067568, -0.094595)
+ mut result := c1.mulinv()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ println(c2.str())
+ println(result.str())
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.12, -0.16)
+ result = c1.mulinv()
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.2, 0.4)
+ result = c1.mulinv()
+ assert result.equals(c2)
+}
+
+fn test_complex_mod() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut result := c1.mod()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert tst_res(result.str(), '8.602325')
+ c1 = cmplx.complex(-3, 4)
+ result = c1.mod()
+ assert result == 5
+ c1 = cmplx.complex(-1, -2)
+ result = c1.mod()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert tst_res(result.str(), '2.236068')
+}
+
+fn test_complex_pow() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-24.0, 70.0)
+ mut result := c1.pow(2)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(117, 44)
+ result = c1.pow(3)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-7, -24)
+ result = c1.pow(4)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_root() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(2.607904, 1.342074)
+ mut result := c1.root(2)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(1.264953, 1.150614)
+ result = c1.root(3)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(1.068059, -0.595482)
+ result = c1.root(4)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_exp() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(111.889015, 97.505457)
+ mut result := c1.exp()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.032543, -0.037679)
+ result = c1.exp()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.153092, -0.334512)
+ result = c1.exp()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_ln() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(2.152033, 0.950547)
+ mut result := c1.ln()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(1.609438, 2.214297)
+ result = c1.ln()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(0.804719, -2.034444)
+ result = c1.ln()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_arg() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(2.152033, 0.950547)
+ mut result := c1.arg()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert tst_res(result.str(), '0.950547')
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(1.609438, 2.214297)
+ result = c1.arg()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert tst_res(result.str(), '2.214297')
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(0.804719, -2.034444)
+ result = c1.arg()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert tst_res(result.str(), '-2.034444')
+}
+
+fn test_complex_log() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut b1 := cmplx.complex(-6, -2)
+ mut c2 := cmplx.complex(0.232873, -1.413175)
+ mut result := c1.log(b1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ b1 = cmplx.complex(3, -1)
+ c2 = cmplx.complex(0.152198, -0.409312)
+ result = c1.log(b1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ b1 = cmplx.complex(0, 9)
+ c2 = cmplx.complex(-0.298243, 1.197981)
+ result = c1.log(b1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_cpow() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut r1 := cmplx.complex(2, 2)
+ mut c2 := cmplx.complex(11.022341, -0.861785)
+ mut result := c1.cpow(r1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ r1 = cmplx.complex(-4, -2)
+ c2 = cmplx.complex(0.118303, 0.063148)
+ result = c1.cpow(r1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ r1 = cmplx.complex(8, -9)
+ c2 = cmplx.complex(-0.000000, 0.000007)
+ result = c1.cpow(r1)
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_sin() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-525.794515, 155.536550)
+ mut result := c1.sin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-3.853738, -27.016813)
+ result = c1.sin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-3.165779, -1.959601)
+ result = c1.sin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_cos() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(155.536809, 525.793641)
+ mut result := c1.cos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-27.034946, 3.851153)
+ result = c1.cos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(2.032723, -3.051898)
+ result = c1.cos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_tan() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-0.000001, 1.000001)
+ mut result := c1.tan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(0.000187, 0.999356)
+ result = c1.tan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.033813, -1.014794)
+ result = c1.tan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_cot() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-0.000001, -0.999999)
+ mut result := c1.cot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(0.000188, -1.000644)
+ result = c1.cot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.032798, 0.984329)
+ result = c1.cot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_sec() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.000517, -0.001749)
+ mut result := c1.sec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.036253, -0.005164)
+ result = c1.sec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(0.151176, 0.226974)
+ result = c1.sec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_csc() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(-0.001749, -0.000517)
+ mut result := c1.csc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.005174, 0.036276)
+ result = c1.csc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.228375, 0.141363)
+ result = c1.csc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_asin() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.617064, 2.846289)
+ mut result := c1.asin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.633984, 2.305509)
+ result = c1.asin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.427079, -1.528571)
+ result = c1.asin()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_acos() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.953732, -2.846289)
+ mut result := c1.acos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(2.204780, -2.305509)
+ result = c1.acos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(1.997875, 1.528571)
+ result = c1.acos()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_atan() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(1.502727, 0.094441)
+ mut result := c1.atan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-1.448307, 0.158997)
+ result = c1.atan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-1.338973, -0.402359)
+ result = c1.atan()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_acot() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.068069, -0.094441)
+ mut result := c1.acot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.122489, -0.158997)
+ result = c1.acot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.231824, 0.402359)
+ result = c1.acot()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_asec() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(1.503480, 0.094668)
+ mut result := c1.asec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(1.689547, 0.160446)
+ result = c1.asec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(1.757114, -0.396568)
+ result = c1.asec()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_acsc() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.067317, -0.094668)
+ mut result := c1.acsc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.118751, -0.160446)
+ result = c1.acsc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.186318, 0.396568)
+ result = c1.acsc()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_sinh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(55.941968, 48.754942)
+ mut result := c1.sinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(6.548120, -7.619232)
+ result = c1.sinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(0.489056, -1.403119)
+ result = c1.sinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_cosh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(55.947047, 48.750515)
+ mut result := c1.cosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-6.580663, 7.581553)
+ result = c1.cosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.642148, 1.068607)
+ result = c1.cosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_tanh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.999988, 0.000090)
+ mut result := c1.tanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-1.000710, 0.004908)
+ result = c1.tanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-1.166736, 0.243458)
+ result = c1.tanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_coth() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(1.000012, -0.000090)
+ mut result := c1.coth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.999267, -0.004901)
+ result = c1.coth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.821330, -0.171384)
+ result = c1.coth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_sech() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.010160, -0.008853)
+ mut result := c1.sech()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.065294, -0.075225)
+ result = c1.sech()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.413149, -0.687527)
+ result = c1.sech()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_csch() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.010159, -0.008854)
+ mut result := c1.csch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(0.064877, 0.075490)
+ result = c1.csch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(0.221501, 0.635494)
+ result = c1.csch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_asinh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(2.844098, 0.947341)
+ mut result := c1.asinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-2.299914, 0.917617)
+ result = c1.asinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-1.469352, -1.063440)
+ result = c1.asinh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_acosh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(2.846289, 0.953732)
+ mut result := c1.acosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(2.305509, 2.204780)
+ result = c1.acosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(1.528571, -1.997875)
+ result = c1.acosh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_atanh() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.067066, 1.476056)
+ mut result := c1.atanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.117501, 1.409921)
+ result = c1.atanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.173287, -1.178097)
+ result = c1.atanh()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_acoth() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.067066, -0.094740)
+ mut result := c1.acoth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.117501, -0.160875)
+ result = c1.acoth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.173287, 0.392699)
+ result = c1.acoth()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+// fn test_complex_asech() {
+// // Tests were also verified on Wolfram Alpha
+// mut c1 := cmplx.complex(5,7)
+// mut c2 := cmplx.complex(0.094668,-1.503480)
+// mut result := c1.asech()
+// // Some issue with precision comparison in f64 using == operator hence serializing to string
+// assert result.str() == c2.str()
+// c1 = cmplx.complex(-3,4)
+// c2 = cmplx.complex(0.160446,-1.689547)
+// result = c1.asech()
+// // Some issue with precision comparison in f64 using == operator hence serializing to string
+// assert result.str() c2.str()
+// c1 = cmplx.complex(-1,-2)
+// c2 = cmplx.complex(0.396568,1.757114)
+// result = c1.asech()
+// // Some issue with precision comparison in f64 using == operator hence serializing to string
+// assert result.str() == c2.str()
+// }
+
+fn test_complex_acsch() {
+ // Tests were also verified on Wolfram Alpha
+ mut c1 := cmplx.complex(5, 7)
+ mut c2 := cmplx.complex(0.067819, -0.094518)
+ mut result := c1.acsch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-3, 4)
+ c2 = cmplx.complex(-0.121246, -0.159507)
+ result = c1.acsch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+ c1 = cmplx.complex(-1, -2)
+ c2 = cmplx.complex(-0.215612, 0.401586)
+ result = c1.acsch()
+ // Some issue with precision comparison in f64 using == operator hence serializing to string
+ assert result.str() == c2.str()
+}
+
+fn test_complex_re_im() {
+ c := cmplx.complex(2.1, 9.05)
+ assert c.re == 2.1
+ assert c.im == 9.05
+}