Case Study 1 — Critical Damping: Why a Defective Matrix Closes the Door

DataField.Dev

Case Study 1 — Critical Damping: Why a Defective Matrix Closes the Door

Field: mechanical and control engineering. This case study cashes in §36.7: a critically-damped system is governed by a defective matrix, and the Jordan block — specifically the $t\,e^{\lambda t}$ term it forces into the matrix exponential — is the mathematical reason such a system settles as fast as physically possible without overshooting. It also previews Chapter 37, where the matrix exponential becomes the solution of every linear system of ODEs.

The problem: settling a system without oscillation or sluggishness

Walk up to a well-designed glass office door, push it, and let go. It swings shut, slows smoothly as it approaches the frame, and closes without slamming and without bouncing back open. The mechanism that achieves this — a hydraulic door-closer — is solving a control problem that recurs everywhere in engineering: bring a system to rest at a target as quickly as possible, with no oscillation and no overshoot. Car suspensions, analog instrument needles, camera autofocus, the read-write head of a hard drive, and the feedback loop of a thermostat all face the same trade-off. Tune the damping too low and the system oscillates around its target, wasting time wobbling. Tune it too high and the system crawls, taking forever to arrive. There is a sharp boundary between these two regimes — and exactly on that boundary, the governing matrix is defective. That is not a coincidence; it is the whole point of this case study.

The physics is the damped harmonic oscillator. A mass on a spring with a damper obeys

$$ x'' + 2c\,x' + k\,x = 0, $$

where $x(t)$ is displacement from equilibrium, $k > 0$ is the spring stiffness (per unit mass), and $c > 0$ is the damping. To bring linear algebra to bear, we convert this second-order scalar equation into a first-order system — the standard move that Chapter 37 will treat in full. Introduce the state vector $\mathbf{x} = (x, x')$ collecting position and velocity. Then $\mathbf{x}' = (x', x'')$, and substituting $x'' = -k x - 2c x'$ gives the linear system $\mathbf{x}' = A\mathbf{x}$ with the companion matrix

$$ A = \begin{bmatrix} 0 & 1 \\ -k & -2c \end{bmatrix}. $$

The behavior of the door is now entirely encoded in the eigenvalues of this $2\times 2$ matrix — exactly the theme that "eigenvalues reveal what a matrix really does," here applied to the destiny of a physical system in time.

Diagnosing the three damping regimes through eigenvalues

The characteristic polynomial of $A$ is $\det(A - \lambda I) = \lambda^2 + 2c\lambda + k$, so the eigenvalues are

$$ \lambda = -c \pm \sqrt{c^2 - k}. $$

Everything depends on the sign of the discriminant $c^2 - k$, and the three cases are the three regimes an engineer designs around:

Underdamped ($c^2 < k$): the eigenvalues are a complex-conjugate pair $-c \pm i\sqrt{k - c^2}$. Complex eigenvalues mean rotation (Chapter 26), and rotation in this context means oscillation — the door swings past the frame and back. The matrix is diagonalizable (distinct eigenvalues), and the solution decays while oscillating, like $e^{-ct}\cos(\omega t)$.
Overdamped ($c^2 > k$): two distinct real negative eigenvalues, $-c \pm \sqrt{c^2 - k}$. The matrix is diagonalizable, and the solution is a sum of two pure exponentials $\alpha e^{\lambda_1 t} + \beta e^{\lambda_2 t}$. No oscillation — but the slower exponential (the eigenvalue closer to zero) dominates, so the door crawls sluggishly to the frame.
Critically damped ($c^2 = k$): the discriminant vanishes and the eigenvalues collide into a single repeated value $\lambda = -c$. This is the boundary, and — as we are about to verify — it is exactly where the matrix becomes defective.

The critically damped case is the engineer's target: it is the unique damping that returns the system to rest in the shortest possible time without overshooting. Any less damping oscillates; any more is slower. And it is the only one of the three regimes whose matrix cannot be diagonalized.

The critical case is defective

Take a concrete critically-damped door-closer with $c = 3$ and $k = 9$ (so $c^2 = 9 = k$). The companion matrix is

$$ A = \begin{bmatrix} 0 & 1 \\ -9 & -6 \end{bmatrix}, $$

with the single repeated eigenvalue $\lambda = -c = -3$ (algebraic multiplicity $2$). Now count eigenvectors:

$$ A - (-3)I = A + 3I = \begin{bmatrix} 3 & 1 \\ -9 & -3 \end{bmatrix}, $$

whose second row is $-3$ times the first — rank $1$. So the eigenspace has dimension $2 - 1 = 1$: geometric multiplicity $1$, short of the algebraic $2$. The critically-damped matrix is defective. It has a single eigenvector — solving $3v_1 + v_2 = 0$ gives $\mathbf{v} = (1, -3)$ — and we are owed one more direction, which we get from a generalized eigenvector exactly as in §36.4.

# Critically damped (c=3, k=9): the companion matrix is defective.
import numpy as np
A = np.array([[0., 1.],
              [-9., -6.]])
vals = np.linalg.eigvals(A)
print("eigenvalues:", np.round(vals, 6))                    # [-3. -3.] (repeated)
print("geometric multiplicity:", 2 - np.linalg.matrix_rank(A + 3*np.eye(2)))  # 1 -> defective

This is the structural fact the eigenvalues alone almost hid: the collision of the two eigenvalues at $-3$ is precisely a collision of modes, and when two modes of a system merge, the matrix that governs them runs short of eigenvectors. The Jordan block is the algebra of that merger.

The matrix exponential, and where the creep comes from

To see the door's actual motion, we solve $\mathbf{x}' = A\mathbf{x}$. The solution — the centerpiece of Chapter 37 — is $\mathbf{x}(t) = e^{At}\mathbf{x}(0)$. For a diagonalizable matrix we would compute $e^{At}$ from the eigenvalues alone, getting pure exponentials. But $A$ is defective, so we need its Jordan form. From the chain (eigenvector $\mathbf{v} = (1,-3)$ and a generalized eigenvector $\mathbf{w}$ solving $(A+3I)\mathbf{w} = \mathbf{v}$), we have $A = PJP^{-1}$ with

$$ J = \begin{bmatrix} -3 & 1 \\ 0 & -3 \end{bmatrix}, \qquad e^{Jt} = e^{-3t}\begin{bmatrix} 1 & t \\ 0 & 1 \end{bmatrix}, $$

using the §36.7 formula for the exponential of a Jordan block. The factor $t$ in the off-diagonal is the fingerprint of the generalized eigenvector. Pulling this back through $P$, every entry of $e^{At}$ — and hence the displacement $x(t)$ — is a combination of $e^{-3t}$ and $t\,e^{-3t}$. With the natural initial condition of a released door ($x(0) = 1$, $x'(0) = 0$), the displacement works out to

$$ x(t) = (1 + 3t)\,e^{-3t}. $$

That $(1 + 3t)$ — a polynomial in $t$ multiplying the exponential — is the signature of defectiveness, the exact analogue of the $k\lambda^{k-1}$ that appears in powers of a Jordan block. A diagonalizable system can only produce pure exponentials; only a defective (critically-damped) system produces this $t\,e^{-3t}$ term. Let us confirm the closed form against scipy's general-purpose matrix exponential, which knows nothing of our hand derivation:

# x(t) = (1 + 3t) e^{-3t} should match expm(At) applied to x(0) = (1, 0).
import numpy as np
from scipy.linalg import expm
A = np.array([[0., 1.], [-9., -6.]])
x0 = np.array([1., 0.])                      # released from rest at displacement 1
for t in [0.2, 0.5, 1.0, 2.0]:
    x_num = (expm(A*t) @ x0)[0]              # displacement from the matrix exponential
    x_cf  = (1 + 3*t) * np.exp(-3*t)         # closed form
    print(f"t={t}: expm={x_num:.6f}, closed form={x_cf:.6f}")
# t=0.2: expm=0.878099, closed form=0.878099
# t=0.5: expm=0.557825, closed form=0.557825
# t=1.0: expm=0.199148, closed form=0.199148
# t=2.0: expm=0.017351, closed form=0.017351

The numbers match to six digits. Read the trajectory physically. At $t = 0$ the door is at displacement $1$. The factor $(1 + 3t)$ initially grows, fighting the decaying $e^{-3t}$, so the door does not simply plunge toward the frame — it eases out, the velocity building before the exponential takes over. By $t = 0.5$ it is at $0.56$, by $t = 1$ at $0.20$, by $t = 2$ down to $0.017$ and effectively closed. Crucially, $x(t) = (1+3t)e^{-3t}$ is positive for all $t > 0$: the door never crosses zero, so it never overshoots — it does not swing past the frame and bounce back. That non-overshooting, smooth approach is exactly the behavior the door-closer is designed for, and it is a direct consequence of the Jordan structure.

Why critical damping wins — and why it is fragile

The $t\,e^{-3t}$ creep is what makes critical damping optimal. Compare it to the overdamped case (more damping, say $c = 5, k = 9$): there the two real eigenvalues are about $-1.15$ and $-7.85$, and the slow one ($-1.15$) dominates, so the door's displacement decays only like $e^{-1.15 t}$ — far slower than the critically-damped $e^{-3t}$. The extra damping, paradoxically, makes the door slower to close, because it drags out the slow mode. Critical damping, sitting exactly at the eigenvalue collision, achieves the fastest non-oscillating decay rate $e^{-ct}$, with the polynomial factor $(1 + ct)$ as the small price the defectiveness exacts. This is why "critically damped" is the gold standard for settling time in control design.

The honest caveat (forward to Chapter 38). Real door-closers, suspensions, and instruments are never exactly critically damped, because $c^2 = k$ is a razor's-edge condition that manufacturing tolerances and temperature changes will perturb. As §36.8 explained, the slightest perturbation splits the repeated eigenvalue and makes the matrix technically diagonalizable (slightly under- or over-damped). Engineers know this: they design near critical damping (often slightly over, to guarantee no overshoot) and treat the exact critical point as an idealization. The Jordan form describes the ideal exactly and the real system approximately — and the $t\,e^{-3t}$ behavior is what you would measure in a system tuned close to critical, even though no physical system sits precisely on the defective point. The mathematics is the truth the engineering approximates.

The takeaway

A critically-damped system is the textbook physical incarnation of a defective matrix. The boundary between oscillation and sluggishness is exactly the boundary where two eigenvalues collide and the governing matrix loses an eigenvector; the missing eigenvector is replaced by a generalized one; and the generalized eigenvector inserts a $t\,e^{\lambda t}$ term into the solution that no diagonalizable system could produce. That term is not a mathematical curiosity — it is the smooth, non-overshooting approach that makes the door close cleanly. When Chapter 37 develops the matrix exponential and phase portraits in full, critical damping will be the canonical example of a degenerate node, the phase-portrait signature of a repeated, defective eigenvalue. The Jordan block, in other words, is the difference between a door that slams and a door that closes.