Notebooks · posts/solving-the-riccati-equation-without-integrating-it

Solving the Riccati equation without integrating it

Adam Zettel · September 23, 2026 · written with an agent

Every LQR controller comes from a quadratic matrix equation, and the tempting way to solve it is to integrate the differential version backwards until it stops moving. On a one-state system that method returns 73.783 where the answer is 414.214, with no warning. The Schur method Laub published in 1979 returns the answer to rounding error, and this post runs it.

A^{⊤} P + P A - P B R^{- 1} B^{⊤} P + W = 0

The gains of a linear-quadratic regulator come out of

A^{⊤} P + P A - P B R^{- 1} B^{⊤} P + W = 0,

quadratic in the matrix $P$ . There is an obvious way to solve it: run $P ˙ = - (A^{⊤} P + P A - P B R^{- 1} B^{⊤} P + W)$ backwards from zero until it stops changing. Each step is one matrix multiply and one matrix add, it converges on most problems, and the earlier LQR post on this site uses it.

Take $A = - 0.001$ , $B = 0.001$ , $W = 1$ , $R = 1$ , one state. The quadratic formula gives $P = 414.214$ exactly. The closed loop it produces decays like $e^{- 0.00141 t}$ , so the Riccati flow settles over roughly seven hundred seconds, and eighty seconds of backward integration gets nowhere near. It returns 73.783, and nothing in its output signals the error.

The fix is to solve a linear eigenvalue problem instead. Stack $A$ and the two weights into the $2 n \times 2 n$ Hamiltonian

H = (A - W - B R^{- 1} B^{⊤} - A^{⊤}),

whose spectrum is symmetric about the imaginary axis: every eigenvalue $λ$ is matched by $- λ$ , so exactly $n$ of the $2 n$ are stable. Take an orthonormal basis $[X_{1}; X_{2}]$ for the invariant subspace of those $n$ , and $P = X_{2} X_{1}^{- 1}$ . No quadratic equation is solved.

Computing that basis is the main work. It is done by reducing $H$ to real Schur form (block upper triangular) and moving the stable blocks to the top left. The animation below shows that reduction: the eleven subdiagonal entries of the Hessenberg matrix, driven to zero by Francis's implicit double-shift QR iteration, one turn of its outer loop at a time. Entry 9 (counting from 0) goes first, and fast. Its size over four successive turns is $3.5 \times 1 0^{- 3}$ , $7.7 \times 1 0^{- 7}$ , $1.9 \times 1 0^{- 13}$ , $3.8 \times 1 0^{- 27}$ : the exponent doubles each time, which is the quadratic convergence a good shift gives. When an entry reaches rounding level, the matrix splits there into two smaller problems and the iteration moves on to the block above. After thirty turns nothing is left to do, and the four entries still nonzero, at $k = 0, 2, 4, 6$ , belong to the 2 by 2 blocks holding the Hamiltonian's four complex-conjugate pairs of eigenvalues.

The last figure compares residuals on ten systems. The Schur method's relative residual stays below $1 0^{- 14}$ on all of them, including those where the largest entry of $P$ runs to five figures and the absolute residual is $5 \times 1 0^{- 7}$ . The matrix sign function, another way to get $P$ from $H$ , is more than thirty times worse on at least one.

Look inside this program

Atlas
The program’s map: every function and which calls which.
There’s no recorded run of this program yet.

Shift+Enter in a cell runs it and moves to the next. Running a cell runs every cell above it too, as one program.

[1]lines 56–394

from lqr import lqr_riccati, lqr_riccati_integrated
from are import (
    are_closed_slowest, are_gains, are_hamiltonian, are_residual, are_residual_rel, are_sign,
    are_solve, are_test_system,
)
from schur import hessenberg, schur_advance, schur_norm

# ---- what integrating it backwards costs, on a system with a slow closed loop
SLOW_A = [[-0.001]]
SLOW_B = [[0.001]]
SLOW_W = [[1.0]]
SLOW_R = 1.0

exact_slow = (-0.001 + sqrt(0.001 * 0.001 + 0.001 * 0.001)) / (0.001 * 0.001)
schur_slow = lqr_riccati(SLOW_A, SLOW_B, SLOW_W, SLOW_R)
naive_slow = lqr_riccati_integrated(SLOW_A, SLOW_B, SLOW_W, SLOW_R)
show exact_slow
show schur_slow
show naive_slow

# the order of the demo system
DEMO_N = 6

#| The system the rest of the post works on: sixth order, one input, unstable
#| to start with, and its $A$ is not symmetric.
def demo_system() {
    A = are_test_system(DEMO_N, 91.0)
    return A
}

DEMO_A = demo_system()
DEMO_B = zeros(DEMO_N, 1)
for i to DEMO_N {
    DEMO_B[i, 0] = 0.4 + 0.3 * mod(i, 2.0)
}
DEMO_R = [[0.7]]
DEMO_W = zeros(DEMO_N, DEMO_N)
for i to DEMO_N {
    DEMO_W[i, i] = 1.0
}

open_re = zeros(DEMO_N)
open_im = zeros(DEMO_N)
eig_general(DEMO_A, open_re, open_im)
worst_open = -1.0e30
for i to DEMO_N {
    worst_open = max(worst_open, open_re[i])
}
show worst_open

DEMO_P = zeros(DEMO_N, DEMO_N)
dim_stable = are_solve(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
DEMO_K = are_gains(DEMO_B, DEMO_R, DEMO_P)
show dim_stable
show are_residual(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
show are_residual_rel(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
show are_closed_slowest(DEMO_A, DEMO_B, DEMO_K)

# ---- the Hamiltonian's spectrum, which is symmetric about the imaginary axis
HAM = are_hamiltonian(DEMO_A, DEMO_B, DEMO_R, DEMO_W)
# the Hamiltonian is 2 DEMO_N square, and has 2 DEMO_N - 1 subdiagonal entries
HAM_N = len(HAM)
NSUB = HAM_N - 1
ham_re = zeros(HAM_N)
ham_im = zeros(HAM_N)
eig_general(HAM, ham_re, ham_im)
ham_trace = 0.0
for i to HAM_N {
    ham_trace = ham_trace + HAM[i, i]
}
show ham_trace

#| Split the Hamiltonian's twelve eigenvalues into the six stable ones the solver
#| keeps and the six unstable ones it discards; each unstable one is the negative
#| of a stable one.
stable_re = zeros(DEMO_N)
stable_im = zeros(DEMO_N)
unstable_re = zeros(DEMO_N)
unstable_im = zeros(DEMO_N)
ns = 0
nu = 0
for i to HAM_N {
    if ham_re[i] < 0.0 {
        stable_re[ns] = ham_re[i]
        stable_im[ns] = ham_im[i]
        ns = ns + 1
    }
    else {
        unstable_re[nu] = ham_re[i]
        unstable_im[nu] = ham_im[i]
        nu = nu + 1
    }
}

figure("The Hamiltonian's twelve eigenvalues", "real part", "imaginary part")
scatter(stable_re, stable_im, "the six stable ones, which P is built from")
scatter(unstable_re, unstable_im, "the six unstable ones, discarded")

# ---- the QR iteration, watched
SWEEPS = 30

#| Run the solver's own outer loop on `H`, one turn at a time, and record the
#| size of every subdiagonal entry after each one. `schur_advance` is the same
#| function `schur` calls; nothing here is a simplified stand-in for it.
def sweep_history(H, turns) {
    n = len(H)
    T = zeros(n, n)
    Q = zeros(n, n)
    junk = hessenberg(H, T, Q)
    anorm = schur_norm(H)
    hist = zeros(turns + 1, n - 1)
    for k to n - 1 {
        hist[0, k] = abs(T[k + 1, k])
    }
    hi = n - 1
    its = 0.0
    for s to turns {
        if hi > 0 {
            nexthi = schur_advance(T, Q, hi, its, anorm)
            if nexthi == hi {
                its = its + 1
            }
            else {
                its = 0.0
            }
            hi = nexthi
        }
        for k to n - 1 {
            hist[s + 1, k] = abs(T[k + 1, k])
        }
    }
    return hist
}

HIST = sweep_history(HAM, SWEEPS)

#| The base-ten logarithm of a magnitude, with a floor so that an entry that has
#| reached exact zero still has somewhere to be drawn.
def log_floor(v) {
    return log(max(v, 1.0e-17)) / 2.302585092994046
}

rows = zeros(NSUB)
for k to NSUB {
    rows[k] = 1.0 * k
}

figure("Each subdiagonal entry shrinking to zero", "turns of the outer loop", "the entry's size")
yscale("log")
sweeps_axis = zeros(SWEEPS + 1)
for s to SWEEPS + 1 {
    sweeps_axis[s] = 1.0 * s
}
last_row = zeros(SWEEPS + 1)
mid_row = zeros(SWEEPS + 1)
top_row = zeros(SWEEPS + 1)
for s to SWEEPS + 1 {
    last_row[s] = HIST[s, 9]
    mid_row[s] = HIST[s, 7]
    top_row[s] = HIST[s, 3]
}
plot(sweeps_axis, last_row, "entry 9, the first to go")
plot(sweeps_axis, mid_row, "entry 7, which goes next")
plot(sweeps_axis, top_row, "entry 3, higher up")

stage("The subdiagonal being driven to zero", -0.6, 10.6, -18.0, 2.5)
axes("which subdiagonal entry", "log10 of its size")
first = zeros(NSUB)
for k to NSUB {
    first[k] = log_floor(HIST[0, k])
}
curve("sub", rows, first)
hline("eps", -16.0)
label("epslabel", "rounding", 8.4, -15.0)
together()
fade_in("sub", 0.3)
fade_in("eps", 0.3)
fade_in("epslabel", 0.3)
apart()
say("Eleven subdiagonal entries, after the Householder reduction to Hessenberg form")
for s to SWEEPS {
    shape = zeros(NSUB)
    for k to NSUB {
        shape[k] = log_floor(HIST[s + 1, k])
    }
    morph("sub", rows, shape, 0.28)
}
say("What is left standing is one entry per 2 by 2 block, and those are the complex pairs")
wait(1.2)

# ---- the Schur method against the sign function, on the same ten systems
NSYS = 10
sizes = zeros(NSYS)
res_schur = zeros(NSYS)
res_sign = zeros(NSYS)
abs_schur = zeros(NSYS)
pmax = zeros(NSYS)
for si to NSYS {
    n = 2 + mod(si, 5.0)
    A = are_test_system(n, 55.0 + 13.0 * si)
    B = zeros(n, 1)
    for i to n {
        B[i, 0] = 0.4 + 0.3 * mod(i, 2.0)
    }
    R = [[0.7]]
    W = zeros(n, n)
    for i to n {
        W[i, i] = 1.0
    }
    Ps = zeros(n, n)
    Pn = zeros(n, n)
    cut = are_solve(A, B, R, W, Ps)
    junk = are_sign(A, B, R, W, Pn)
    sizes[si] = 1.0 * si
    res_schur[si] = log_floor(are_residual_rel(A, B, R, W, Ps))
    abs_schur[si] = are_residual(A, B, R, W, Ps)
    res_sign[si] = log_floor(are_residual_rel(A, B, R, W, Pn))
    big = 0.0
    for i to n {
        for j to n {
            big = max(big, abs(Ps[i, j]))
        }
    }
    pmax[si] = big
}
show pmax

figure("Relative residual, ten systems, two methods", "system", "$\\log_{10}$ of the relative residual")
plot(sizes, res_schur, "the Schur method")
plot(sizes, res_sign, "the matrix sign function")
style("the Schur method", "s1 noline circle")
style("the matrix sign function", "s2 noline square")

test "the backward integration is wrong by four fifths on the slow system" {
    assert_near(exact_slow, 414.2135623730951, 1.0e-9)
    assert_lt(abs(schur_slow[0, 0] - exact_slow) / exact_slow, 1.0e-14)
    assert_near(naive_slow[0, 0], 73.783, 1.0e-3)
    assert_lt(0.8, abs(naive_slow[0, 0] - exact_slow) / exact_slow)
}

test "the demo system starts unstable and the gains put every mode in the left half plane" {
    assert_lt(0.0, worst_open)
    assert_lt(are_closed_slowest(DEMO_A, DEMO_B, DEMO_K), -1.0e-8)
}

test "the Hamiltonian has exactly six stable eigenvalues and zero trace" {
    assert_near(dim_stable, 6.0, 1.0e-12)
    assert_near(ham_trace, 0.0, 1.0e-12)
    count = 0.0
    for i to HAM_N {
        if ham_re[i] < 0.0 {
            count = count + 1.0
        }
    }
    assert_near(count, 6.0, 1.0e-12)
}

test "every eigenvalue of the Hamiltonian is matched by minus itself" {
    for i to HAM_N {
        near = 1.0e30
        for j to HAM_N {
            near = min(near, abs(ham_re[i] + ham_re[j]) + abs(ham_im[i] - ham_im[j]))
        }
        assert_lt(near, 1.0e-9)
    }
}

test "the largest P, five figures, still has a relative residual below 1e-14" {
    # system 9: largest entry 22253.8, absolute residual 5.34e-7, measured
    assert_lt(10000.0, pmax[9])
    assert_lt(pmax[9], 100000.0)
    assert_lt(res_schur[9], -14.0)
    assert_lt(4.5e-7, abs_schur[9])
    assert_lt(abs_schur[9], 5.5e-7)
}

test "P solves the Riccati equation with a relative residual below 1e-14" {
    assert_lt(are_residual_rel(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P), 1.0e-14)
    for i to DEMO_N {
        for j to DEMO_N {
            assert_near(DEMO_P[i, j], DEMO_P[j, i], 1.0e-12)
        }
    }
}

test "thirty turns leaves four subdiagonal entries, one per complex pair" {
    standing = 0.0
    for k to NSUB {
        if HIST[SWEEPS, k] != 0.0 {
            standing = standing + 1.0
        }
    }
    assert_near(standing, 4.0, 1.0e-12)
    # and they are the ones at the head of each 2 by 2 block
    assert_lt(0.1, HIST[SWEEPS, 0])
    assert_lt(0.1, HIST[SWEEPS, 2])
    assert_lt(0.1, HIST[SWEEPS, 4])
    assert_lt(0.1, HIST[SWEEPS, 6])
    assert_near(HIST[SWEEPS, 1], 0.0, 1.0e-300)
    assert_near(HIST[SWEEPS, 8], 0.0, 1.0e-300)
}

test "entry 9 converges quadratically: the exponent doubles each turn" {
    a = HIST[3, 9]
    b = HIST[4, 9]
    c = HIST[5, 9]
    d = HIST[6, 9]
    assert_near(a, 3.4939e-3, 1.0e-6)
    assert_near(b, 7.70091e-7, 1.0e-11)
    assert_near(c, 1.85967e-13, 1.0e-17)
    assert_near(d, 3.76417e-27, 1.0e-31)
    assert_lt(d, 1.0e-25)
    # each step roughly doubles how many digits are gone
    assert_lt(1.7, log(b) / log(a))
    assert_lt(1.7, log(c) / log(b))
}

test "the Schur method beats the sign function on every one of the ten systems" {
    for si to NSYS {
        assert_lt(res_schur[si], -14.0)
    }
    worse = 0.0
    for si to NSYS {
        worse = max(worse, res_sign[si] - res_schur[si])
    }
    # somewhere in there the sign function is two orders of magnitude behind
    assert_lt(1.5, worse)
}

test "some of these systems are nearly uncontrollable, which is the hard case" {
    biggest = 0.0
    for si to NSYS {
        biggest = max(biggest, pmax[si])
    }
    assert_lt(1000.0, biggest)
}

# SPDX-License-Identifier: MIT OR Apache-2.0
# SPDX-FileCopyrightText: 2026 The Viv authors

from lqr import lqr_riccati, lqr_riccati_integrated
from are import (
    are_closed_slowest, are_gains, are_hamiltonian, are_residual, are_residual_rel, are_sign,
    are_solve, are_test_system,
)
from schur import hessenberg, schur_advance, schur_norm

# ---- what integrating it backwards costs, on a system with a slow closed loop
SLOW_A = [[-0.001]]
SLOW_B = [[0.001]]
SLOW_W = [[1.0]]
SLOW_R = 1.0

exact_slow = (-0.001 + sqrt(0.001 * 0.001 + 0.001 * 0.001)) / (0.001 * 0.001)
schur_slow = lqr_riccati(SLOW_A, SLOW_B, SLOW_W, SLOW_R)
naive_slow = lqr_riccati_integrated(SLOW_A, SLOW_B, SLOW_W, SLOW_R)
show exact_slow
show schur_slow
show naive_slow

# the order of the demo system
DEMO_N = 6

#| The system the rest of the post works on: sixth order, one input, unstable
#| to start with, and its $A$ is not symmetric.
def demo_system() {
    A = are_test_system(DEMO_N, 91.0)
    return A
}

DEMO_A = demo_system()
DEMO_B = zeros(DEMO_N, 1)
for i to DEMO_N {
    DEMO_B[i, 0] = 0.4 + 0.3 * mod(i, 2.0)
}
DEMO_R = [[0.7]]
DEMO_W = zeros(DEMO_N, DEMO_N)
for i to DEMO_N {
    DEMO_W[i, i] = 1.0
}

open_re = zeros(DEMO_N)
open_im = zeros(DEMO_N)
eig_general(DEMO_A, open_re, open_im)
worst_open = -1.0e30
for i to DEMO_N {
    worst_open = max(worst_open, open_re[i])
}
show worst_open

DEMO_P = zeros(DEMO_N, DEMO_N)
dim_stable = are_solve(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
DEMO_K = are_gains(DEMO_B, DEMO_R, DEMO_P)
show dim_stable
show are_residual(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
show are_residual_rel(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P)
show are_closed_slowest(DEMO_A, DEMO_B, DEMO_K)

# ---- the Hamiltonian's spectrum, which is symmetric about the imaginary axis
HAM = are_hamiltonian(DEMO_A, DEMO_B, DEMO_R, DEMO_W)
# the Hamiltonian is 2 DEMO_N square, and has 2 DEMO_N - 1 subdiagonal entries
HAM_N = len(HAM)
NSUB = HAM_N - 1
ham_re = zeros(HAM_N)
ham_im = zeros(HAM_N)
eig_general(HAM, ham_re, ham_im)
ham_trace = 0.0
for i to HAM_N {
    ham_trace = ham_trace + HAM[i, i]
}
show ham_trace

#| Split the Hamiltonian's twelve eigenvalues into the six stable ones the solver
#| keeps and the six unstable ones it discards; each unstable one is the negative
#| of a stable one.
stable_re = zeros(DEMO_N)
stable_im = zeros(DEMO_N)
unstable_re = zeros(DEMO_N)
unstable_im = zeros(DEMO_N)
ns = 0
nu = 0
for i to HAM_N {
    if ham_re[i] < 0.0 {
        stable_re[ns] = ham_re[i]
        stable_im[ns] = ham_im[i]
        ns = ns + 1
    }
    else {
        unstable_re[nu] = ham_re[i]
        unstable_im[nu] = ham_im[i]
        nu = nu + 1
    }
}

figure("The Hamiltonian's twelve eigenvalues", "real part", "imaginary part")
scatter(stable_re, stable_im, "the six stable ones, which P is built from")
scatter(unstable_re, unstable_im, "the six unstable ones, discarded")

# ---- the QR iteration, watched
SWEEPS = 30

#| Run the solver's own outer loop on `H`, one turn at a time, and record the
#| size of every subdiagonal entry after each one. `schur_advance` is the same
#| function `schur` calls; nothing here is a simplified stand-in for it.
def sweep_history(H, turns) {
    n = len(H)
    T = zeros(n, n)
    Q = zeros(n, n)
    junk = hessenberg(H, T, Q)
    anorm = schur_norm(H)
    hist = zeros(turns + 1, n - 1)
    for k to n - 1 {
        hist[0, k] = abs(T[k + 1, k])
    }
    hi = n - 1
    its = 0.0
    for s to turns {
        if hi > 0 {
            nexthi = schur_advance(T, Q, hi, its, anorm)
            if nexthi == hi {
                its = its + 1
            }
            else {
                its = 0.0
            }
            hi = nexthi
        }
        for k to n - 1 {
            hist[s + 1, k] = abs(T[k + 1, k])
        }
    }
    return hist
}

HIST = sweep_history(HAM, SWEEPS)

#| The base-ten logarithm of a magnitude, with a floor so that an entry that has
#| reached exact zero still has somewhere to be drawn.
def log_floor(v) {
    return log(max(v, 1.0e-17)) / 2.302585092994046
}

rows = zeros(NSUB)
for k to NSUB {
    rows[k] = 1.0 * k
}

figure("Each subdiagonal entry shrinking to zero", "turns of the outer loop", "the entry's size")
yscale("log")
sweeps_axis = zeros(SWEEPS + 1)
for s to SWEEPS + 1 {
    sweeps_axis[s] = 1.0 * s
}
last_row = zeros(SWEEPS + 1)
mid_row = zeros(SWEEPS + 1)
top_row = zeros(SWEEPS + 1)
for s to SWEEPS + 1 {
    last_row[s] = HIST[s, 9]
    mid_row[s] = HIST[s, 7]
    top_row[s] = HIST[s, 3]
}
plot(sweeps_axis, last_row, "entry 9, the first to go")
plot(sweeps_axis, mid_row, "entry 7, which goes next")
plot(sweeps_axis, top_row, "entry 3, higher up")

stage("The subdiagonal being driven to zero", -0.6, 10.6, -18.0, 2.5)
axes("which subdiagonal entry", "log10 of its size")
first = zeros(NSUB)
for k to NSUB {
    first[k] = log_floor(HIST[0, k])
}
curve("sub", rows, first)
hline("eps", -16.0)
label("epslabel", "rounding", 8.4, -15.0)
together()
fade_in("sub", 0.3)
fade_in("eps", 0.3)
fade_in("epslabel", 0.3)
apart()
say("Eleven subdiagonal entries, after the Householder reduction to Hessenberg form")
for s to SWEEPS {
    shape = zeros(NSUB)
    for k to NSUB {
        shape[k] = log_floor(HIST[s + 1, k])
    }
    morph("sub", rows, shape, 0.28)
}
say("What is left standing is one entry per 2 by 2 block, and those are the complex pairs")
wait(1.2)

# ---- the Schur method against the sign function, on the same ten systems
NSYS = 10
sizes = zeros(NSYS)
res_schur = zeros(NSYS)
res_sign = zeros(NSYS)
abs_schur = zeros(NSYS)
pmax = zeros(NSYS)
for si to NSYS {
    n = 2 + mod(si, 5.0)
    A = are_test_system(n, 55.0 + 13.0 * si)
    B = zeros(n, 1)
    for i to n {
        B[i, 0] = 0.4 + 0.3 * mod(i, 2.0)
    }
    R = [[0.7]]
    W = zeros(n, n)
    for i to n {
        W[i, i] = 1.0
    }
    Ps = zeros(n, n)
    Pn = zeros(n, n)
    cut = are_solve(A, B, R, W, Ps)
    junk = are_sign(A, B, R, W, Pn)
    sizes[si] = 1.0 * si
    res_schur[si] = log_floor(are_residual_rel(A, B, R, W, Ps))
    abs_schur[si] = are_residual(A, B, R, W, Ps)
    res_sign[si] = log_floor(are_residual_rel(A, B, R, W, Pn))
    big = 0.0
    for i to n {
        for j to n {
            big = max(big, abs(Ps[i, j]))
        }
    }
    pmax[si] = big
}
show pmax

figure("Relative residual, ten systems, two methods", "system", "$\\log_{10}$ of the relative residual")
plot(sizes, res_schur, "the Schur method")
plot(sizes, res_sign, "the matrix sign function")
style("the Schur method", "s1 noline circle")
style("the matrix sign function", "s2 noline square")

test "the backward integration is wrong by four fifths on the slow system" {
    assert_near(exact_slow, 414.2135623730951, 1.0e-9)
    assert_lt(abs(schur_slow[0, 0] - exact_slow) / exact_slow, 1.0e-14)
    assert_near(naive_slow[0, 0], 73.783, 1.0e-3)
    assert_lt(0.8, abs(naive_slow[0, 0] - exact_slow) / exact_slow)
}

test "the demo system starts unstable and the gains put every mode in the left half plane" {
    assert_lt(0.0, worst_open)
    assert_lt(are_closed_slowest(DEMO_A, DEMO_B, DEMO_K), -1.0e-8)
}

test "the Hamiltonian has exactly six stable eigenvalues and zero trace" {
    assert_near(dim_stable, 6.0, 1.0e-12)
    assert_near(ham_trace, 0.0, 1.0e-12)
    count = 0.0
    for i to HAM_N {
        if ham_re[i] < 0.0 {
            count = count + 1.0
        }
    }
    assert_near(count, 6.0, 1.0e-12)
}

test "every eigenvalue of the Hamiltonian is matched by minus itself" {
    for i to HAM_N {
        near = 1.0e30
        for j to HAM_N {
            near = min(near, abs(ham_re[i] + ham_re[j]) + abs(ham_im[i] - ham_im[j]))
        }
        assert_lt(near, 1.0e-9)
    }
}

test "the largest P, five figures, still has a relative residual below 1e-14" {
    # system 9: largest entry 22253.8, absolute residual 5.34e-7, measured
    assert_lt(10000.0, pmax[9])
    assert_lt(pmax[9], 100000.0)
    assert_lt(res_schur[9], -14.0)
    assert_lt(4.5e-7, abs_schur[9])
    assert_lt(abs_schur[9], 5.5e-7)
}

test "P solves the Riccati equation with a relative residual below 1e-14" {
    assert_lt(are_residual_rel(DEMO_A, DEMO_B, DEMO_R, DEMO_W, DEMO_P), 1.0e-14)
    for i to DEMO_N {
        for j to DEMO_N {
            assert_near(DEMO_P[i, j], DEMO_P[j, i], 1.0e-12)
        }
    }
}

test "thirty turns leaves four subdiagonal entries, one per complex pair" {
    standing = 0.0
    for k to NSUB {
        if HIST[SWEEPS, k] != 0.0 {
            standing = standing + 1.0
        }
    }
    assert_near(standing, 4.0, 1.0e-12)
    # and they are the ones at the head of each 2 by 2 block
    assert_lt(0.1, HIST[SWEEPS, 0])
    assert_lt(0.1, HIST[SWEEPS, 2])
    assert_lt(0.1, HIST[SWEEPS, 4])
    assert_lt(0.1, HIST[SWEEPS, 6])
    assert_near(HIST[SWEEPS, 1], 0.0, 1.0e-300)
    assert_near(HIST[SWEEPS, 8], 0.0, 1.0e-300)
}

test "entry 9 converges quadratically: the exponent doubles each turn" {
    a = HIST[3, 9]
    b = HIST[4, 9]
    c = HIST[5, 9]
    d = HIST[6, 9]
    assert_near(a, 3.4939e-3, 1.0e-6)
    assert_near(b, 7.70091e-7, 1.0e-11)
    assert_near(c, 1.85967e-13, 1.0e-17)
    assert_near(d, 3.76417e-27, 1.0e-31)
    assert_lt(d, 1.0e-25)
    # each step roughly doubles how many digits are gone
    assert_lt(1.7, log(b) / log(a))
    assert_lt(1.7, log(c) / log(b))
}

test "the Schur method beats the sign function on every one of the ten systems" {
    for si to NSYS {
        assert_lt(res_schur[si], -14.0)
    }
    worse = 0.0
    for si to NSYS {
        worse = max(worse, res_sign[si] - res_schur[si])
    }
    # somewhere in there the sign function is two orders of magnitude behind
    assert_lt(1.5, worse)
}

test "some of these systems are nearly uncontrollable, which is the hard case" {
    biggest = 0.0
    for si to NSYS {
        biggest = max(biggest, pmax[si])
    }
    assert_lt(1000.0, biggest)
}

# SPDX-License-Identifier: MIT OR Apache-2.0
# SPDX-FileCopyrightText: 2026 The Viv authors