ruda-solver 0.1.0

Explicit scientific solvers: real/complex/sparse host algebra and opt-in small batched device kernels
Documentation 
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

// SPDX-License-Identifier: Apache-2.0
//! Host complex FP64 LU, Hermitian Cholesky and thin Householder QR.
//! Each component is f64 (128 storage bits per complex value); not CUDA Complex64.
use crate::{SolverError,Tolerance};
use std::ops::{Add,Sub,Mul,Div,Neg};
#[derive(Clone,Copy,Debug,Default,PartialEq)]
pub struct Complex64 {pub re:f64,pub im:f64}
impl Complex64 {
    pub const ZERO:Self=Self{re:0.0,im:0.0}; pub const ONE:Self=Self{re:1.0,im:0.0};
    pub const fn new(re:f64,im:f64)->Self{Self{re,im}}
    pub fn conj(self)->Self{Self::new(self.re,-self.im)}
    pub fn abs(self)->f64{self.re.hypot(self.im)}
    pub fn is_finite(self)->bool{self.re.is_finite()&&self.im.is_finite()}
    pub fn scale(self,s:f64)->Self{Self::new(self.re*s,self.im*s)}
}
impl Add for Complex64 {type Output=Self;fn add(self,b:Self)->Self{Self::new(self.re+b.re,self.im+b.im)}}
impl Sub for Complex64 {type Output=Self;fn sub(self,b:Self)->Self{Self::new(self.re-b.re,self.im-b.im)}}
impl Neg for Complex64 {type Output=Self;fn neg(self)->Self{Self::new(-self.re,-self.im)}}
impl Mul for Complex64 {type Output=Self;fn mul(self,b:Self)->Self{Self::new(self.re.mul_add(b.re,-self.im*b.im),self.re.mul_add(b.im,self.im*b.re))}}
impl Div for Complex64 {
    type Output=Self;
    fn div(self,b:Self)->Self {
        // Scaled denominator avoids squaring |b|. Algorithms check output finiteness.
        let s=b.re.abs().max(b.im.abs()); let br=b.re/s;let bi=b.im/s;
        let d=br*br+bi*bi;let ar=self.re/s;let ai=self.im/s;
        Self::new((ar*br+ai*bi)/d,(ai*br-ar*bi)/d)
    }
}
fn check(x:Complex64)->Result<Complex64,SolverError>{if x.is_finite(){Ok(x)}else{Err(SolverError::Arithmetic("complex intermediate"))}}
#[derive(Clone,Debug,PartialEq)]
pub struct ComplexMatrix {pub(crate) rows:usize,pub(crate) cols:usize,pub(crate) data:Vec<Complex64>}
impl ComplexMatrix {
    pub fn new(rows:usize,cols:usize,data:Vec<Complex64>)->Result<Self,SolverError>{
        if rows==0||cols==0{return Err(SolverError::Shape("empty complex matrix"));}
        let n=rows.checked_mul(cols).ok_or(SolverError::SizeOverflow)?;
        if n!=data.len(){return Err(SolverError::Shape("complex matrix length"));}
        for(i,x)in data.iter().enumerate(){if !x.is_finite(){return Err(SolverError::NonFinite{index:i});}}
        Ok(Self{rows,cols,data})
    }
    pub fn zeros(rows:usize,cols:usize)->Result<Self,SolverError>{
        let n=rows.checked_mul(cols).filter(|&n|n<=isize::MAX as usize/16).ok_or(SolverError::SizeOverflow)?;
        let mut v=Vec::new();v.try_reserve_exact(n).map_err(|_|SolverError::Allocation)?;v.resize(n,Complex64::ZERO);
        Self::new(rows,cols,v)
    }
    pub fn identity(n:usize)->Result<Self,SolverError>{let mut a=Self::zeros(n,n)?;for i in 0..n{a.data[i*n+i]=Complex64::ONE;}Ok(a)}
    pub fn rows(&self)->usize{self.rows} pub fn columns(&self)->usize{self.cols}
    pub fn values(&self)->&[Complex64]{&self.data}
    pub fn adjoint(&self)->Result<Self,SolverError>{let mut b=Self::zeros(self.cols,self.rows)?;for i in 0..self.rows{for j in 0..self.cols{b.data[j*self.rows+i]=self.data[i*self.cols+j].conj();}}Ok(b)}
    fn square(&self)->Result<usize,SolverError>{if self.rows!=self.cols{Err(SolverError::Shape("complex square matrix required"))}else{Ok(self.rows)}}
    fn max_abs(&self)->f64{self.data.iter().fold(0.0f64,|s,x|s.max(x.abs()))}
}
#[derive(Clone,Debug)]
pub struct ComplexLu {packed:ComplexMatrix,pivots:Vec<usize>}
impl ComplexLu {
    /// P A = L U; partial row pivoting, no input mutation.
    pub fn factor(a:&ComplexMatrix,tolerance:Tolerance)->Result<Self,SolverError>{
        let n=a.square()?;let cutoff=tolerance.threshold(a.max_abs())?;let mut packed=a.clone();let mut pivots=Vec::new();
        for k in 0..n{
            let mut p=k;for i in k+1..n{if packed.data[i*n+k].abs()>packed.data[p*n+k].abs(){p=i;}}
            let pivot=packed.data[p*n+k].abs();if pivot<=cutoff{return Err(SolverError::Singular{index:k,pivot,threshold:cutoff});}
            pivots.push(p);if p!=k{for j in 0..n{packed.data.swap(k*n+j,p*n+j);}}
            for i in k+1..n{let r=check(packed.data[i*n+k]/packed.data[k*n+k])?;packed.data[i*n+k]=r;
                for j in k+1..n{packed.data[i*n+j]=check(packed.data[i*n+j]-r*packed.data[k*n+j])?;}}
        }Ok(Self{packed,pivots})
    }
    pub fn packed(&self)->&ComplexMatrix{&self.packed} pub fn pivots(&self)->&[usize]{&self.pivots}
    pub fn solve(&self,b:&ComplexMatrix)->Result<ComplexMatrix,SolverError>{self.solve_impl(b,false)}
    pub fn solve_adjoint(&self,b:&ComplexMatrix)->Result<ComplexMatrix,SolverError>{self.solve_impl(b,true)}
    fn solve_impl(&self,b:&ComplexMatrix,adjoint:bool)->Result<ComplexMatrix,SolverError>{
        let n=self.packed.rows;let r=b.cols;if b.rows!=n{return Err(SolverError::Shape("complex LU RHS"));}let mut x=b.clone();
        if !adjoint{
            for(k,&p)in self.pivots.iter().enumerate(){for c in 0..r{x.data.swap(k*r+c,p*r+c);}}
            for i in 0..n{for c in 0..r{let mut s=x.data[i*r+c];for j in 0..i{s=check(s-self.packed.data[i*n+j]*x.data[j*r+c])?;}x.data[i*r+c]=s;}}
            for i in(0..n).rev(){for c in 0..r{let mut s=x.data[i*r+c];for j in i+1..n{s=check(s-self.packed.data[i*n+j]*x.data[j*r+c])?;}x.data[i*r+c]=check(s/self.packed.data[i*n+i])?;}}
        }else{
            for i in 0..n{for c in 0..r{let mut s=x.data[i*r+c];for j in 0..i{s=check(s-self.packed.data[j*n+i].conj()*x.data[j*r+c])?;}x.data[i*r+c]=check(s/self.packed.data[i*n+i].conj())?;}}
            for i in(0..n).rev(){for c in 0..r{let mut s=x.data[i*r+c];for j in i+1..n{s=check(s-self.packed.data[j*n+i].conj()*x.data[j*r+c])?;}x.data[i*r+c]=s;}}
            for k in(0..n).rev(){for c in 0..r{x.data.swap(k*r+c,self.pivots[k]*r+c);}}
        }Ok(x)
    }
}
#[derive(Clone,Debug)]pub struct ComplexCholesky{lower:ComplexMatrix}
impl ComplexCholesky{
    /// A=L L^H. Lower triangle defines the matrix after Hermitian validation.
    pub fn factor(a:&ComplexMatrix,tolerance:Tolerance)->Result<Self,SolverError>{
        let n=a.square()?;tolerance.validate()?;let cutoff=tolerance.threshold(a.max_abs())?;
        for i in 0..n{if a.data[i*n+i].im.abs()>cutoff{return Err(SolverError::NotSymmetric{row:i,column:i});}
            for j in 0..i{let x=a.data[i*n+j];let y=a.data[j*n+i].conj();let s=x.abs().max(y.abs());
                if s>0.0&&(x.scale(1.0/s)-y.scale(1.0/s)).abs()>(tolerance.absolute/s).max(tolerance.relative){return Err(SolverError::NotSymmetric{row:i,column:j});}}}
        let mut l=ComplexMatrix::zeros(n,n)?;
        for j in 0..n{
            let mut d=a.data[j*n+j].re;for k in 0..j{let z=l.data[j*n+k];d-=z.re*z.re+z.im*z.im;}
            if !d.is_finite(){return Err(SolverError::Arithmetic("complex Cholesky pivot"));}
            if d<=0.0{return Err(SolverError::NotPositiveDefinite{index:j,pivot:d});}
            l.data[j*n+j]=Complex64::new(d.sqrt(),0.0);
            for i in j+1..n{let mut s=a.data[i*n+j];for k in 0..j{s=check(s-l.data[i*n+k]*l.data[j*n+k].conj())?;}
                l.data[i*n+j]=check(s/l.data[j*n+j])?;}
        }Ok(Self{lower:l})
    }
    pub fn lower(&self)->&ComplexMatrix{&self.lower}
    pub fn solve(&self,b:&ComplexMatrix)->Result<ComplexMatrix,SolverError>{
        let(n,r)=(self.lower.rows,b.cols);if b.rows!=n{return Err(SolverError::Shape("complex Cholesky RHS"));}let mut x=b.clone();
        for i in 0..n{for c in 0..r{let mut s=x.data[i*r+c];for j in 0..i{s=check(s-self.lower.data[i*n+j]*x.data[j*r+c])?;}x.data[i*r+c]=check(s/self.lower.data[i*n+i])?;}}
        for i in(0..n).rev(){for c in 0..r{let mut s=x.data[i*r+c];for j in i+1..n{s=check(s-self.lower.data[j*n+i].conj()*x.data[j*r+c])?;}x.data[i*r+c]=check(s/self.lower.data[i*n+i])?;}}Ok(x)
    }
}
#[derive(Clone,Debug)]pub struct ComplexQr{q:ComplexMatrix,r:ComplexMatrix,rank:usize}
impl ComplexQr{
    /// Unpivoted thin Householder QR, m>=n. Two-dimensional host data only.
    /// Rank is a diagonal diagnostic, not a rank-revealing decomposition.
    pub fn factor(a:&ComplexMatrix,tolerance:Tolerance)->Result<Self,SolverError>{
        let(m,n)=(a.rows,a.cols);if m<n{return Err(SolverError::Shape("complex QR requires m>=n"));}
        let cutoff=tolerance.threshold(a.max_abs())?;let mut r=a.clone();
        let mut reflectors:Vec<Vec<Complex64>>=Vec::new();
        for k in 0..n{
            let norm=crate::numerics::norm_iter((k..m).map(|i|r.data[i*n+k].abs()))?;
            let mut v=vec![Complex64::ZERO;m-k];
            if norm>0.0{
                let x0=r.data[k*n+k];let phase=if x0.abs()==0.0{Complex64::ONE}else{x0.scale(1.0/x0.abs())};
                for i in k..m{v[i-k]=r.data[i*n+k].scale(1.0/norm);}v[0]=v[0]+phase;
                let vn=crate::numerics::norm_iter(v.iter().map(|z|z.abs()))?;for z in &mut v{*z=z.scale(1.0/vn);}
                for j in k..n{
                    let mut d=Complex64::ZERO;for i in k..m{d=check(d+v[i-k].conj()*r.data[i*n+j])?;}
                    d=d.scale(2.0);for i in k..m{r.data[i*n+j]=check(r.data[i*n+j]-v[i-k]*d)?;}
                }
                for i in k+1..m{r.data[i*n+k]=Complex64::ZERO;}
            }reflectors.push(v);
        }
        let mut q=ComplexMatrix::zeros(m,n)?;for j in 0..n{q.data[j*n+j]=Complex64::ONE;}
        for k in(0..n).rev(){let v=&reflectors[k];for j in 0..n{
            let mut d=Complex64::ZERO;for i in k..m{d=check(d+v[i-k].conj()*q.data[i*n+j])?;}
            d=d.scale(2.0);for i in k..m{q.data[i*n+j]=check(q.data[i*n+j]-v[i-k]*d)?;}
        }}
        let mut thin=ComplexMatrix::zeros(n,n)?;for i in 0..n{for j in i..n{thin.data[i*n+j]=r.data[i*n+j];}}
        let rank=(0..n).filter(|&j|thin.data[j*n+j].abs()>cutoff).count();Ok(Self{q,r:thin,rank})
    }
    pub fn q(&self)->&ComplexMatrix{&self.q} pub fn r(&self)->&ComplexMatrix{&self.r}pub fn rank(&self)->usize{self.rank}
    pub fn least_squares(&self,b:&ComplexMatrix)->Result<ComplexMatrix,SolverError>{
        let(m,n,p)=(self.q.rows,self.q.cols,b.cols);if b.rows!=m{return Err(SolverError::Shape("complex QR RHS"));}
        if self.rank<n{return Err(SolverError::RankDeficient{rank:self.rank,columns:n});}
        let mut x=ComplexMatrix::zeros(n,p)?;for j in 0..n{for c in 0..p{
            let mut s=Complex64::ZERO;for i in 0..m{s=check(s+self.q.data[i*n+j].conj()*b.data[i*p+c])?;}x.data[j*p+c]=s;
        }}for i in(0..n).rev(){for c in 0..p{let mut s=x.data[i*p+c];for j in i+1..n{s=check(s-self.r.data[i*n+j]*x.data[j*p+c])?;}x.data[i*p+c]=check(s/self.r.data[i*n+i])?;}}Ok(x)
    }
}