ripopt 0.7.1

A memory-safe interior point optimizer 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

//! Issue #7 reproducer: dual infeasibility stalls at ~1e-3 on exp/log objectives
//!
//! Reproduces the exact problem from the issue, matching discopt's formulation:
//! - DENSE Jacobian and Hessian (PyO3 bindings use m*n and n*(n+1)/2)
//! - '>=' constraints normalized to '<=' with negated body (Expression.__ge__)
//! - Constraint bodies compiled as `body - rhs` with rhs=0
//!
//! discopt formulation of constraints:
//!   x1+x2 >= 3  →  body = 3-(x1+x2), sense <=, bounds [-inf, 0]
//!   x1 <= 4*y1  →  body = x1-4*y1, sense <=, bounds [-inf, 0]
//!   x2 <= 4*y2  →  body = x2-4*y2, sense <=, bounds [-inf, 0]
//!   y1+y2 <= 1+y3 → body = y1+y2-1-y3, sense <=, bounds [-inf, 0]

use ripopt::{NlpProblem, SolveStatus, SolverOptions};

struct ExpActivation {
    x_l: [f64; 5],
    x_u: [f64; 5],
    x0: [f64; 5],
}

impl ExpActivation {
    fn relaxation() -> Self {
        Self {
            x_l: [0.0, 0.0, 0.0, 0.0, 0.0],
            x_u: [4.0, 4.0, 1.0, 1.0, 1.0],
            x0: [2.0, 2.0, 0.5, 0.5, 0.5],
        }
    }

    fn branch(fixes: &[(usize, f64)], parent_x: &[f64; 5]) -> Self {
        let mut x_l = [0.0, 0.0, 0.0, 0.0, 0.0];
        let mut x_u = [4.0, 4.0, 1.0, 1.0, 1.0];
        for &(idx, val) in fixes {
            x_l[idx] = val;
            x_u[idx] = val;
        }
        let mut x0 = *parent_x;
        for i in 0..5 {
            x0[i] = x0[i].clamp(x_l[i], x_u[i]);
        }
        Self { x_l, x_u, x0 }
    }
}

impl NlpProblem for ExpActivation {
    fn num_variables(&self) -> usize { 5 }
    fn num_constraints(&self) -> usize { 4 }

    fn bounds(&self, x_l: &mut [f64], x_u: &mut [f64]) {
        x_l.copy_from_slice(&self.x_l);
        x_u.copy_from_slice(&self.x_u);
    }

    fn constraint_bounds(&self, g_l: &mut [f64], g_u: &mut [f64]) {
        // All constraints normalized to <= 0 (discopt style)
        for i in 0..4 {
            g_l[i] = f64::NEG_INFINITY;
            g_u[i] = 0.0;
        }
    }

    fn initial_point(&self, x0: &mut [f64]) {
        x0.copy_from_slice(&self.x0);
    }

    fn objective(&self, x: &[f64], _new_x: bool, obj: &mut f64) -> bool {
        *obj = x[0].exp() + x[1].exp() + 5.0 * x[2] + 6.0 * x[3] + 8.0 * x[4];
        true
    }

    fn gradient(&self, x: &[f64], _new_x: bool, grad: &mut [f64]) -> bool {
        grad[0] = x[0].exp();
        grad[1] = x[1].exp();
        grad[2] = 5.0;
        grad[3] = 6.0;
        grad[4] = 8.0;
        true
    }

    fn constraints(&self, x: &[f64], _new_x: bool, g: &mut [f64]) -> bool {
        // discopt normalization: >= becomes negated <=
        g[0] = 3.0 - x[0] - x[1];            // 3-(x1+x2) <= 0
        g[1] = x[0] - 4.0 * x[2];             // x1-4*y1 <= 0
        g[2] = x[1] - 4.0 * x[3];             // x2-4*y2 <= 0
        g[3] = x[2] + x[3] - 1.0 - x[4];      // y1+y2-1-y3 <= 0
        true
    }

    // Dense Jacobian: 4*5 = 20 entries (matching PyO3 bindings)
    fn jacobian_structure(&self) -> (Vec<usize>, Vec<usize>) {
        let mut rows = Vec::with_capacity(20);
        let mut cols = Vec::with_capacity(20);
        for i in 0..4 {
            for j in 0..5 {
                rows.push(i);
                cols.push(j);
            }
        }
        (rows, cols)
    }

    fn jacobian_values(&self, _x: &[f64], _new_x: bool, vals: &mut [f64]) -> bool {
        vals.fill(0.0);
        // g0 = 3-x1-x2: dg/dx1=-1, dg/dx2=-1
        vals[0 * 5 + 0] = -1.0;
        vals[0 * 5 + 1] = -1.0;
        // g1 = x1-4*y1: dg/dx1=1, dg/dy1=-4
        vals[1 * 5 + 0] = 1.0;
        vals[1 * 5 + 2] = -4.0;
        // g2 = x2-4*y2: dg/dx2=1, dg/dy2=-4
        vals[2 * 5 + 1] = 1.0;
        vals[2 * 5 + 3] = -4.0;
        // g3 = y1+y2-1-y3: dg/dy1=1, dg/dy2=1, dg/dy3=-1
        vals[3 * 5 + 2] = 1.0;
        vals[3 * 5 + 3] = 1.0;
        vals[3 * 5 + 4] = -1.0;
        true
    }

    // Dense lower-triangle Hessian: 5*(5+1)/2 = 15 entries
    fn hessian_structure(&self) -> (Vec<usize>, Vec<usize>) {
        let mut rows = Vec::with_capacity(15);
        let mut cols = Vec::with_capacity(15);
        for i in 0..5 {
            for j in 0..=i {
                rows.push(i);
                cols.push(j);
            }
        }
        (rows, cols)
    }

    fn hessian_values(&self, x: &[f64], _new_x: bool, obj_factor: f64, _lambda: &[f64], vals: &mut [f64]) -> bool {
        vals.fill(0.0);
        vals[0] = obj_factor * x[0].exp();  // H[0,0]
        vals[2] = obj_factor * x[1].exp();  // H[1,1]
        true
    }
}

fn run_problem(name: &str, problem: &ExpActivation, opts: &SolverOptions) -> (SolveStatus, usize, f64, [f64; 5]) {
    println!("\n=== {} ===", name);
    let result = ripopt::solve(problem, opts);
    println!("  Status:     {:?}", result.status);
    println!("  Objective:  {:.10e}", result.objective);
    println!("  Solution:   x=[{:.4}, {:.4}] y=[{:.4}, {:.4}, {:.4}]",
        result.x[0], result.x[1], result.x[2], result.x[3], result.x[4]);
    println!("  Iterations: {}", result.iterations);
    if result.status != SolveStatus::Optimal {
        println!("  >>> NON-OPTIMAL STATUS");
    }
    let mut sol = [0.0; 5];
    sol.copy_from_slice(&result.x[..5]);
    (result.status, result.iterations, result.objective, sol)
}

fn main() {
    let opts = SolverOptions {
        print_level: 5,
        tol: 1e-7,
        max_iter: 3000,
        // Disable fallback solvers to expose the raw IPM stall
        enable_lbfgs_fallback: false,
        enable_al_fallback: false,
        enable_sqp_fallback: false,
        enable_slack_fallback: false,
        enable_lbfgs_hessian_fallback: false,
        disable_nlp_restoration: true,
        ..SolverOptions::default()
    };

    // Node 1: Root relaxation
    let (s1, i1, _, root_sol) = run_problem("Node 1: Root relaxation",
        &ExpActivation::relaxation(), &opts);

    // Node 2: y1=1
    let (s2, i2, _, sol2) = run_problem("Node 2: y1=1",
        &ExpActivation::branch(&[(2, 1.0)], &root_sol), &opts);

    // Node 3: y1=0
    let (s3, i3, _, sol3) = run_problem("Node 3: y1=0",
        &ExpActivation::branch(&[(2, 0.0)], &root_sol), &opts);

    // Node 4: y1=1, y2=1
    let (s4, i4, _, _) = run_problem("Node 4: y1=1, y2=1",
        &ExpActivation::branch(&[(2, 1.0), (3, 1.0)], &sol2), &opts);

    // Node 5: y1=1, y2=0
    let (s5, i5, _, _) = run_problem("Node 5: y1=1, y2=0",
        &ExpActivation::branch(&[(2, 1.0), (3, 0.0)], &sol2), &opts);

    // Node 6: y1=0, y2=1
    let (s6, i6, _, sol6) = run_problem("Node 6: y1=0, y2=1",
        &ExpActivation::branch(&[(2, 0.0), (3, 1.0)], &sol3), &opts);

    // Node 7: y1=0, y2=1, y3=0
    let (s7, i7, _, _) = run_problem("Node 7: y1=0, y2=1, y3=0",
        &ExpActivation::branch(&[(2, 0.0), (3, 1.0), (4, 0.0)], &sol6), &opts);

    // Summary
    println!("\n=== SUMMARY ===");
    let results = vec![
        ("Root", s1, i1),
        ("y1=1", s2, i2),
        ("y1=0", s3, i3),
        ("y1=1,y2=1", s4, i4),
        ("y1=1,y2=0", s5, i5),
        ("y1=0,y2=1", s6, i6),
        ("y1=0,y2=1,y3=0", s7, i7),
    ];
    let mut any_stall = false;
    for (name, status, iters) in &results {
        let flag = if *status != SolveStatus::Optimal { " *** STALL" } else { "" };
        if *status != SolveStatus::Optimal { any_stall = true; }
        println!("  {:25} {:?} ({} iters){}", name, status, iters, flag);
    }
    if any_stall {
        println!("\nREPRODUCED: dual stall on exp() B&B subproblems");
    } else {
        println!("\nNOT REPRODUCED");
    }
}