diffsol 0.16.2

A library for solving ordinary differential equations (ODEs) in Rust.
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
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

use std::{cell::RefCell, marker::PhantomData};

use crate::{
    Matrix, NonLinearOp, NonLinearOpAdjoint, NonLinearOpJacobian, NonLinearOpSens,
    NonLinearOpSensAdjoint, Op, Scale, Vector,
};
use num_traits::{One, Zero};
use std::ops::MulAssign;

use super::{BuilderOp, OpStatistics, ParameterisedOp};

/// An [`Op`] wrapping a user-provided closure with `std::autodiff` support.
///
/// When the `autodiff` feature is enabled, [`call_func`](Self::call_func) is annotated
/// with `#[autodiff_forward]` and `#[autodiff_reverse]` to automatically generate
/// `call_jvp`, `call_vjp`, and `call_sens_vjp`.
pub struct ClosureAutodiff<M: Matrix, F> {
    func: F,
    nstates: usize,
    nout: usize,
    nparams: usize,
    statistics: RefCell<OpStatistics>,
    tmp_nstates: RefCell<M::V>,
    tmp_nstates2: RefCell<M::V>,
    ctx: M::C,
    _phantom: PhantomData<M>,
}

impl<M: Matrix, F> ClosureAutodiff<M, F> {
    pub fn new(func: F, nstates: usize, nout: usize, nparams: usize, ctx: M::C) -> Self {
        Self {
            func,
            nstates,
            nout,
            nparams,
            tmp_nstates: RefCell::new(M::V::zeros(nstates, ctx.clone())),
            tmp_nstates2: RefCell::new(M::V::zeros(nstates, ctx.clone())),
            statistics: RefCell::new(OpStatistics::default()),
            ctx,
            _phantom: PhantomData,
        }
    }
}

impl<M: Matrix, F> Op for ClosureAutodiff<M, F> {
    type V = M::V;
    type T = M::T;
    type M = M;
    type C = M::C;
    fn nstates(&self) -> usize {
        self.nstates
    }
    fn nout(&self) -> usize {
        self.nout
    }
    fn nparams(&self) -> usize {
        self.nparams
    }
    fn statistics(&self) -> OpStatistics {
        self.statistics.borrow().clone()
    }
    fn context(&self) -> &Self::C {
        &self.ctx
    }
}

impl<M: Matrix, F> BuilderOp for ClosureAutodiff<M, F> {
    fn calculate_sparsity(&mut self, _y0: &Self::V, _t0: Self::T, _p: &Self::V) {}
    fn set_nstates(&mut self, nstates: usize) {
        self.nstates = nstates;
        self.tmp_nstates = RefCell::new(M::V::zeros(nstates, self.ctx.clone()));
        self.tmp_nstates2 = RefCell::new(M::V::zeros(nstates, self.ctx.clone()));
    }
    fn set_nout(&mut self, nout: usize) {
        self.nout = nout;
    }
    fn set_nparams(&mut self, nparams: usize) {
        self.nparams = nparams;
    }
}

#[cfg(feature = "autodiff")]
mod autodiff_impl {
    use super::*;
    use crate::Vector;
    use std::autodiff::{autodiff_forward, autodiff_reverse};

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> ClosureAutodiff<M, F> {
        #[autodiff_forward(call_jvp, Const, Dual, Const, Const, Dual)]
        #[autodiff_forward(call_sens_jvp, Const, Const, Dual, Const, Dual)]
        #[autodiff_reverse(call_vjp, Const, Duplicated, Const, Const, Duplicated)]
        #[autodiff_reverse(call_sens_vjp, Const, Const, Duplicated, Const, Duplicated)]
        pub fn call_func(&self, x: &M::V, p: &M::V, t: M::T, y: &mut M::V) {
            (self.func)(x, p, t, y)
        }
    }

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> NonLinearOp
        for ParameterisedOp<'_, ClosureAutodiff<M, F>>
    {
        fn call_inplace(&self, x: &M::V, t: M::T, y: &mut M::V) {
            self.op.statistics.borrow_mut().increment_call();
            self.op.call_func(x, self.p, t, y);
        }
    }

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> NonLinearOpJacobian
        for ParameterisedOp<'_, ClosureAutodiff<M, F>>
    {
        fn jac_mul_inplace(&self, x: &M::V, t: M::T, v: &M::V, y: &mut M::V) {
            self.op.statistics.borrow_mut().increment_jac_mul();
            let mut tmp_nstates = self.op.tmp_nstates.borrow_mut();
            self.op.call_jvp(x, v, self.p, t, &mut tmp_nstates, y);
        }
        fn jacobian_inplace(&self, x: &Self::V, t: Self::T, y: &mut Self::M) {
            self.op.statistics.borrow_mut().increment_matrix();
            self._default_jacobian_inplace(x, t, y);
        }
        fn jacobian_sparsity(&self) -> Option<<Self::M as Matrix>::Sparsity> {
            None
        }
    }

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> NonLinearOpSens
        for ParameterisedOp<'_, ClosureAutodiff<M, F>>
    {
        fn sens_mul_inplace(&self, x: &M::V, t: M::T, v: &M::V, y: &mut M::V) {
            let mut tmp_nstates = self.op.tmp_nstates.borrow_mut();
            self.op.call_sens_jvp(x, self.p, v, t, &mut tmp_nstates, y);
        }
    }

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> NonLinearOpAdjoint
        for ParameterisedOp<'_, ClosureAutodiff<M, F>>
    {
        fn jac_transpose_mul_inplace(&self, x: &Self::V, t: Self::T, v: &Self::V, y: &mut Self::V) {
            self.op.statistics.borrow_mut().increment_jac_adj_mul();
            let mut tmp_nstates = self.op.tmp_nstates.borrow_mut();
            let mut tmp_nstates2 = self.op.tmp_nstates2.borrow_mut();
            tmp_nstates.copy_from(v);
            y.fill(M::T::zero());
            self.op
                .call_vjp(x, y, self.p, t, &mut tmp_nstates2, &mut tmp_nstates);
            y.mul_assign(Scale(-M::T::one()));
        }
        fn adjoint_inplace(&self, x: &Self::V, t: Self::T, y: &mut Self::M) {
            self._default_adjoint_inplace(x, t, y);
        }
        fn adjoint_sparsity(&self) -> Option<<Self::M as Matrix>::Sparsity> {
            None
        }
    }

    impl<M: Matrix, F: Fn(&M::V, &M::V, M::T, &mut M::V)> NonLinearOpSensAdjoint
        for ParameterisedOp<'_, ClosureAutodiff<M, F>>
    {
        fn sens_transpose_mul_inplace(
            &self,
            x: &Self::V,
            t: Self::T,
            v: &Self::V,
            y: &mut Self::V,
        ) {
            let mut tmp_nstates = self.op.tmp_nstates.borrow_mut();
            let mut tmp_nstates2 = self.op.tmp_nstates2.borrow_mut();
            tmp_nstates.copy_from(&v);
            y.fill(M::T::zero());
            self.op
                .call_sens_vjp(x, self.p, y, t, &mut tmp_nstates2, &mut tmp_nstates);
            y.mul_assign(Scale(-M::T::one()));
        }
        fn sens_adjoint_inplace(&self, x: &Self::V, t: Self::T, y: &mut Self::M) {
            self._default_sens_adjoint_inplace(x, t, y);
        }
        fn sens_adjoint_sparsity(&self) -> Option<<Self::M as Matrix>::Sparsity> {
            None
        }
    }
}

#[cfg(test)]
mod tests {
    use crate::{
        context::nalgebra::NalgebraContext, NalgebraMat, NalgebraVec, NonLinearOp,
        NonLinearOpAdjoint, NonLinearOpJacobian, NonLinearOpSens, NonLinearOpSensAdjoint,
        ParameterisedOp, Vector,
    };

    use super::ClosureAutodiff;

    type M = NalgebraMat<f64>;
    type V = NalgebraVec<f64>;

    fn nonlinear(x: &V, p: &V, t: f64, y: &mut V) {
        y[0] = p[0] * x[0] * x[0] + 2.0 * x[1] + t;
        y[1] = 3.0 * x[0] + p[1] * x[1] - t;
    }

    #[test]
    fn autodiff_closure_generates_state_and_parameter_gradients() {
        let ctx = NalgebraContext::default();
        let op = ClosureAutodiff::<M, _>::new(nonlinear, 2, 2, 2, ctx);
        let p = V::from_vec(vec![4.0, 5.0], ctx);
        let pop = ParameterisedOp::new(&op, &p);
        let x = V::from_vec(vec![2.0, 3.0], ctx);

        let mut value = V::zeros(2, ctx);
        pop.call_inplace(&x, 0.5, &mut value);
        value.assert_eq_st(&V::from_vec(vec![22.5, 20.5], ctx), 1e-12);

        let direction = V::from_vec(vec![7.0, 11.0], ctx);
        let mut jvp = V::zeros(2, ctx);
        pop.jac_mul_inplace(&x, 0.5, &direction, &mut jvp);
        jvp.assert_eq_st(&V::from_vec(vec![134.0, 76.0], ctx), 1e-12);

        let mut parameter_jvp = V::zeros(2, ctx);
        pop.sens_mul_inplace(&x, 0.5, &direction, &mut parameter_jvp);
        parameter_jvp.assert_eq_st(&V::from_vec(vec![28.0, 33.0], ctx), 1e-12);

        let mut state_vjp = V::zeros(2, ctx);
        pop.jac_transpose_mul_inplace(&x, 0.5, &direction, &mut state_vjp);
        state_vjp.assert_eq_st(&V::from_vec(vec![-145.0, -69.0], ctx), 1e-12);

        let mut parameter_vjp = V::zeros(2, ctx);
        pop.sens_transpose_mul_inplace(&x, 0.5, &direction, &mut parameter_vjp);
        parameter_vjp.assert_eq_st(&V::from_vec(vec![-28.0, -33.0], ctx), 1e-12);
    }
}