1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
//!
//!Traits for metric properties and function
//!

use crate::algebra::*;
use crate::analysis::*;

///
///A real-valued function on a set `X` that quantifies the "distance" between objects
///
///This is rigorously defined as a function `d:X⨯X -> R` such that
/// * `d(x,y) > 0` for all `x != y`
/// * `d(x,x) = 0` for all `x`
/// * `d(x,z) <= d(x,y) + d(y,z)` for all `x`,`y`, and `z`
///
/// ## Implementation Note
///
///Due to the fact that there are often _many_ metrics for any given space and that a given
///kind of metric often applies to a whole family of spaces, this trait has been written
///with the intent of being implemented on type _other_ than the set `X` to which it applies.
///This way, a construction can implemented only once while still having access to all applicable
///metrics.
///
///A good example of this would a struct implementing all L<sup>p</sup> norms for a single
///implementation of a given ℝ<sup>3</sup> or similar
///
///
pub trait Metric<X, R:Real> {
    fn distance(&self, x1:X, x2:X) -> R;
}

///
///A real-valued function from a ring module that quantifies it's length, allowing for null-vectors
///
///Specifically, a seminorm ‖•‖ is a function from a ring module `X` over `K` to the reals such that:
/// * `K` has a seminorm |•|
/// * `‖x‖ >= 0`
/// * `‖cx‖ = |c|‖x‖`
/// * `‖x+y‖ <= ‖x‖ + ‖y‖`
///
///This is distinct from a NormedMetric in that it is allowed to be 0 for non-zero vectors
///
pub trait Seminorm<K:UnitalRing, X:RingModule<K>, R:Real> {
    #[inline] fn norm(&self, x:X) -> R;
    #[inline] fn normalize(&self, x:X) -> X where K:From<R> {x.clone() * K::from(self.norm(x).inv())}
}

///
///A real-valued on a ring-module that quantifies it's length, disallowing null-vectors
///
///Specifically, a norm ‖x‖ is a function from a ring module `X` over `K` to the reals such that:
/// * `K` has a norm |•|
/// * `‖x‖ > 0` for all `x != 0`
/// * `‖cx‖ = |c|‖x‖`
/// * `‖x+y‖ <= ‖x‖ + ‖y‖`
///
///This is distinct from a [Seminorm] in that it is _not_ allowed to be 0 for non-zero vectors
///
pub trait Norm<K:UnitalRing, X:RingModule<K>, R:Real>: Seminorm<K,X,R> {}

///
///A hermitian form with the added restriction that `x•x` be Real and `x•x>0` for nonzero `x`
///
///Because of this restriction, any inner product automatically also defines a [Norm] as the square
///root of any element with itself
///
pub trait InnerProduct<K:ComplexRing, M:RingModule<K>>: HermitianForm<K,M> + Norm<K,M,K::Real>{}

///A metric that also defines a norm
pub trait NormedMetric<K,X,R> = Norm<K,X,R> + Metric<X,R> where K:UnitalRing, X:RingModule<K>, R:Real;

///
///A metric on vector-spaces using the [inner product](InnerProductSpace) of two vectors
///
///For finite dimensional real-vector spaces, this is simply the Euclidean metric, and for functions
///on measure-spaces, this gives the L2-metric
///
#[derive(Copy, Clone, PartialEq, Eq, Hash, Debug)]
pub struct InnerProductMetric;

impl<R:Real, V:InnerProductSpace<R>> Metric<V,R> for InnerProductMetric {
    #[inline(always)] fn distance(&self, x1:V, x2:V) -> R {x1.dist_euclid(x2)}
}

impl<K:ComplexRing, V:InnerProductSpace<K>> Seminorm<K,V,K::Real> for InnerProductMetric {
    #[inline(always)] fn norm(&self, x:V) -> K::Real {x.norm()}
}

impl<K:ComplexRing, V:InnerProductSpace<K>> Norm<K,V,K::Real> for InnerProductMetric {}

impl<K:ComplexRing, V:InnerProductSpace<K>> SesquilinearForm<K,V> for InnerProductMetric {
    #[inline] fn product_of(&self, v1:V, v2:V) -> K {v1.inner_product(v2)}
    #[inline] fn sigma(&self, x:K) -> K {x.conj()}
    #[inline] fn sigma_inv(&self, x:K) -> K {x.conj()}
}

impl<K:ComplexRing, V:InnerProductSpace<K>> ReflexiveForm<K,V> for InnerProductMetric {}
impl<K:ComplexRing, V:InnerProductSpace<K>> SymSesquilinearForm<K,V> for InnerProductMetric {}
impl<K:Real, V:InnerProductSpace<K>> BilinearForm<K,V> for InnerProductMetric {}

impl<K:ComplexRing, V:InnerProductSpace<K>> ComplexSesquilinearForm<K,V> for InnerProductMetric {}
impl<K:ComplexRing, V:InnerProductSpace<K>> InnerProduct<K,V> for InnerProductMetric {}

///
///A ring module over some complex ring with an inner product
///
///Rigorously, this operation is defined as a binary product `⟨,⟩:MxM->ℂ` from a module over a complex
///ring to the complex numbers such that:
/// * `⟨x+y,z⟩ = ⟨x,z⟩ + ⟨y,z⟩`
/// * `⟨x•y⟩ = ̅⟨̅y̅,̅x̅⟩`
/// * `⟨c*x,y⟩ = c*⟨x•y⟩`
/// * `⟨x,x⟩` is real-valued and `⟨x,x⟩ > 0` whenenever `x != 0`
///
///Practically, however, the inner-product is simply a way to combine together the ideas of
///(Euclidean) vector lengths, orthogonality/perpendicularity, and vector projections into one operation
///Specifically:
/// * Vector length becomes: `‖x‖ = √⟨x,x⟩`
/// * `x` and `y` are orthogonal iff `⟨x,y⟩ == 0`
/// * The component of `x` in the "direction" of `y` is quantified by `⟨x,y⟩*y/⟨y,y⟩`
///
///# Uniqueness
///
///While this trait treats the inner-product as a property of the space it acts upon, it is
///worth noting that _technically_, for all non-trivial inner-product spaces,
///there are *infinitely* many possible choices for it's output.
///In fact, another inner-product can always be found by simply taking the pre-existing operation
///and scaling it by some non-zero scalar. _However_, despite this, the choice of product is almost always
///taken as intrinsic to to whichever space being considered, since the pick of product is entirely
///determined by it's value on the space's basis vectors which themselves are usually core to that
///vector's representation.
///
///However, if a space with more options is desired, the option to use the [SesquilinearForm]
///system in addition or in lieu of this trait also exists.
///
///# Examples
///
/// * The standard N-dimensional real-valued dot product: `⟨x,y⟩ = x₁*y₁ + ... xₙ*yₙ`
/// * The N-dimensional complex dot product: `⟨x,y⟩ = x₁*̅y̅₁ + ... xₙ*̅y̅ₙ`
/// * The L₂ inner-product between complex-valued functions: `⟨f,g⟩ = ∫f(x)*̅g(x)dx`
///
///
pub trait InnerProductSpace<F: ComplexRing>: RingModule<F> {

    ///
    ///A binary scalar operation `⟨,⟩:MxM->ℂ` between two module element such that:
    /// * `⟨x+y,z⟩ = ⟨x,z⟩ + ⟨y,z⟩`
    /// * `⟨x•y⟩ = ̅⟨̅y̅,̅x̅⟩`
    /// * `⟨c*x,y⟩ = c*⟨x•y⟩`
    /// * `⟨x,x⟩` is real-valued and `⟨x,x⟩ > 0` whenenever `x != 0`
    ///
    fn inner_product(self, rhs:Self) -> F;

    ///
    ///The square of the norm induced by the inner-product
    ///
    ///This is equivalent to `⟨x,x⟩` for all `x`
    ///
    #[inline] fn norm_sqrd(self) -> F::Real {self.clone().inner_product(self).as_real()}

    ///
    ///The norm induced by the inner-product
    ///
    ///This is equivalent to `√⟨x,x⟩` for all `x`
    ///
    #[inline] fn norm(self) -> F::Real {self.norm_sqrd().sqrt()}

    ///The distance between two elements as defined by `√⟨x-y,x-y⟩`
    #[inline] fn dist_euclid(self, rhs: Self) -> F::Real {(self - rhs).norm()}

    ///
    ///Divides an element by its length. Only available if the reals are fully contained in this module's scalars
    ///
    ///Note that this can panic or return an error value if self if is zero
    ///
    #[inline] fn normalized(self) -> Self where F:From<<F as ComplexSubset>::Real> {
        self.clone() * F::from(self.norm().inv())
    }

    ///Determines if two elements are normal to each other. Equivalent to `⟨x,y⟩ == 0`
    #[inline] fn orthogonal(self, rhs:Self) -> bool {self.inner_product(rhs).is_zero()}

    ///Computes the orthogonal component of `rhs` to `self`. equivalent to `rhs - self.project(rhs)`
    #[inline] fn reject(self, rhs: Self) -> Self where F:ComplexField { rhs.clone() - self.project(rhs) }

    ///Computes the parallel component of `rhs` to `self`. equivalent to `⟨self,rhs⟩*rhs/⟨self,self⟩`
    #[inline] fn project(self, rhs: Self) -> Self where F:ComplexField {
        let l = self.clone().inner_product(self.clone()).inv() * self.clone().inner_product(rhs);
        self * l
    }

    ///
    ///Computes a measure of the angle between two module elements
    ///
    ///For finite real-spaces, this works exactly as expected, but for infinite or complex spaces,
    ///the particular interpretation varies depending on context.
    ///
    ///Rigorously, this method is equivalent equivalent to `acos(⟨self,rhs⟩/(‖self‖*‖rhs‖))`, and is
    ///based upon the finite real result that `⟨x,y⟩ = ‖x‖*‖y‖*cos(θ)`
    ///
    #[inline] fn angle(self, rhs: Self) -> F where F:Trig+From<<F as ComplexSubset>::Real> {
        let l1 = self.clone().norm();
        let l2 = rhs.clone().norm();
        (self.inner_product(rhs) * (l1*l2).inv().into()).acos()
    }
}

#[cfg(feature = "std")]
macro_rules! impl_metric {
    (@int $($f:ident)*) => {$(
        impl InnerProductSpace<$f> for $f {
            #[inline(always)] fn inner_product(self, rhs:Self) -> $f {self * rhs}
            #[inline(always)] fn norm(self) -> <$f as ComplexSubset>::Real {$f::as_real(self.abs())}
            #[inline(always)] fn orthogonal(self, rhs:Self) -> bool {self==0 || rhs==0}
        }
    )*};
    (@float $($f:ident)*) => {$(
        impl InnerProductSpace<$f> for $f {

            #[inline(always)] fn inner_product(self, rhs:Self) -> $f {self * rhs}
            #[inline(always)] fn norm(self) -> $f {self.abs()}
            #[inline(always)] fn normalized(self) -> $f {self.signum()}

            #[inline(always)] fn orthogonal(self, rhs:Self) -> bool {self==0.0 || rhs==0.0}
            #[inline(always)] fn reject(self, rhs: Self) -> Self { if self==0.0 {rhs} else {0.0} }
            #[inline(always)] fn project(self, rhs: Self) -> Self { if self==0.0 {0.0} else {rhs} }

            #[inline(always)]
            fn angle(self, rhs: Self) -> $f {
                if (self==0.0) ^ (rhs==0.0) {
                    ::core::$f::consts::FRAC_PI_2
                } else if (self<0.0) ^ (rhs<0.0) {
                    ::core::$f::consts::PI
                } else {
                    0.0
                }
            }
        }
    )*}
}

// Necessary do to issue #60021
#[cfg(feature = "std")]
mod impls {
    use super::{ ComplexSubset, InnerProductSpace };
    impl_metric!(@float f32 f64);
    impl_metric!(@int i32 i64);
}