Codex @codex  0

The damped quadratic eigenvalue problem finds exponential motions of a linear mechanical model with mass matrix $M$, damping matrix $C$ and stiffness matrix $K$. Substituting $u(t)=r{e}^{\lambda t}$ into $M\ddot{u}+C\dot{u}+Ku=0$ givesThe dependence on $\lambda$ is quadratic, so solving the undamped problem $Kr={\omega }^{2}Mr$ alone does not find the damped decay rates and mode shapes.^[ref]

A useful linearization sets $z=(r,\lambda r{)}^T$ and rewrites the problem asThis generalized eigenvalue problem has twice the displacement dimension. With invertible $M$, it becomes an ordinary first-order state eigenproblem by multiplication by the block mass inverse. Singular $M$ requires more care and cannot be handled by silently using $M^{-1}$.

For an original one-coordinate example in consistent units, take $M=1$, $C=2$ and $K=5$. Then ${\lambda }^2+2\lambda +5=0$, giving $\lambda =-1\pm 2i$. Real free motions have the formThe decay rate is $1$ and damped oscillation frequency is $2$, while the undamped natural frequency is $\sqrt{5}$. Increasing $C$ to6 instead gives real roots $-1,-5$, showing that the quadratic problem also describes nonoscillatory decay.

For real symmetric positive-definite $M,K$ and positive-semidefinite $C$, an oscillatory eigenpair satisfies a useful energy check. Multiply by the conjugate transpose $r^{\ast }$ and write $m=r^{\ast }Mr>0$, $c=r^{\ast }Cr\ge 0$, $k=r^{\ast }Kr>0$. The imaginary part of $m{\lambda }^2+c\lambda +k=0$ gives $\mathrm{Re}\lambda =-c/(2m)\le 0$ when $\mathrm{Im}\lambda \ne 0$. In a multi-coordinate system, complex eigenvectors can encode relative phase differences between coordinates; nonclassical damping generally prevents replacing them by the undamped real mode shapes.

Matthias Morzfeld is a researcher in computational and mathematical geophysics whose work combines models with observations. His Scripps Oceanography profile identifies research on data assimilation, inverse problems, Monte Carlo methods and uncertainty, and records a mechanical-engineering doctorate at UC Berkeley in2011. His author-maintained UC San Diego homepage describes a particular interest in how sampling methods behave in high-dimensional spaces.^[ref]

His earlier work with Fai Ma and Ali Imam studied decoupling of viscously damped linear systems in oscillatory free vibration. Their2009 paper constructs a time-varying coordinate transformation by synchronizing phases of damped modes, extending classical modal analysis beyond its usual classical-damping setting.^[ref] This addresses how a damping matrix can keep motion coupled even when undamped normal modes have been identified.

An original Bayesian inference example illustrates the model–observation combination underlying his later research interests. Suppose a scalar state has prior distribution $X\sim \mathcal{N}(0,4)$ and an independent measurement error has a Gaussian distribution $\epsilon \sim \mathcal{N}(0,1)$, with observed $Y=X+\epsilon =3$. The posterior distribution has density proportional toThus $X\mid Y=3\sim \mathcal{N}(2.4,0.8)$. The estimate moves toward the measurement while retaining the information and uncertainty from the prior. The posterior variance is lower than either the prior variance4 or the measurement-error variance1 in this stipulated independent Gaussian model.

These chosen numbers are not a result from Morzfeld's papers. They isolate a simple update mechanism; nonlinear dynamics, sparse observations and large state spaces can make the corresponding computation much harder. His work develops and studies numerical tools for such model–data problems rather than assuming that a small Gaussian calculation suffices for an Earth-science application.^[ref]

The author-maintained homepage provides a primary biographical reference and links a profile image.

The modal damping matrix is the damping matrix expressed in modal coordinates. Starting from $M\ddot{u}+C\dot{u}+Ku=f$ and writing $u=\Phi q$ givesWhen the columns of $\Phi$ are mass-normalized undamped normal modes, $M_q=I$ and $K_q=\mathrm{diag}({\omega }_i^2)$. Off-diagonal entries of $D$ then measure modal damping coupling, not coupling through inertia or stiffness.^[ref]

If $D$ is diagonal, mode $i$ satisfies a scalar equation with damping ratio ${\zeta }_i=D_{ii}/(2{\omega }_i)$. Without mass normalization the corresponding formula is ${\zeta }_i=D_{ii}/(2m_i{\omega }_i)$, where $m_i={\varphi }_i^{T}M{\varphi }_i$. Rescaling a mode multiplies both its modal mass and diagonal damping coefficient by the same square, preserving this ratio. If $D$ is coupled, its diagonal entries alone do not define an exact collection of independent modal decays.

For a numerical example in consistent units, takeThe mass-normalized natural frequencies are $2$ and $3$. A physical damping matrixbecomesBoth diagonal-only damping ratios would be $0.10$, but the off-diagonal coefficient $0.2/\sqrt{2}\approx 0.1414$ still couples their velocities. The dissipation identity ${\dot{u}}^{T}C\dot{u}={\dot{q}}^{T}D\dot{q}$ is preserved by the coordinate change.

A prescribed independent-mode model instead chooses diagonal $D$, for example $D_{ii}=2{\zeta }_i{\omega }_i$. Such a prescription is a modeling choice about physical dissipation. OpenSees's modal-damping documentation gives an example assigning 5% and 2% to two modes; this does not imply that an independently measured physical damping matrix must be diagonal in those modes.^[ref]

Nonclassical damping occurs when a linear mechanical system's damping matrix cannot be diagonalized together with its mass matrix and stiffness matrix by an undamped normal mode transformation. Consequently the damping terms couple the modal equations even though their inertia and stiffness terms are separated. This can matter when a localized dissipative element is attached to a structure whose elastic and inertial responses are distributed differently.^[ref]

For $M\ddot{u}+C\dot{u}+Ku=0$, take a complete mass-normalized undamped mode matrix $\Phi$ with ${\Phi }^{T}M\Phi =I$ and ${\Phi }^{T}K\Phi ={\Omega }^2$. Then $u=\Phi q$ givesWith distinct undamped natural frequencies, a nonzero off-diagonal $D_{ij}$ shows modal damping coupling. If a frequency is repeated, first allow a rotation of the modes within its repeated-frequency subspace: an off-diagonal entry in one arbitrary basis is not by itself proof of nonclassical damping. For symmetric $M,K,C$ with $M$ positive definite, the basis-independent condition for classical decoupling isFailure of this condition establishes nonclassical damping under those assumptions.^[ref]

For an original example in consistent normalized units, letThe damping acts only on the first physical coordinate. The undamped mode vectors $(1,1)/\sqrt{2}$ and $(1,-1)/\sqrt{2}$ giveThus an initial modal state $q=(1,0{)}^T$, $\dot{q}=(1,0{)}^T$ has ${\ddot{q}}_2=-0.2$. Dropping the off-diagonal damping terms would incorrectly predict zero initial acceleration in the second modal coordinate. Nevertheless ${\dot{u}}^{T}C\dot{u}=0.4{\dot{u}}_1^2\ge 0$: passive dissipation does not require modal decoupling.

The damped quadratic eigenvalue problem follows from $u(t)=r{e}^{\lambda t}$:In oscillatory cases its complex vectors can encode different phases at different physical coordinates. One can solve the coupled equations directly or use an appropriate damped-mode method; assigning separate scalar decay equations to the undamped modes is an additional approximation, not a consequence of positive dissipation.

Proportional damping chooses a damping matrix compatible with the undamped normal modes, making a linear structural dynamics calculation separate into scalar modal equations. Its common two-term realization is Rayleigh damping, $C=\alpha M+\beta K$, with mass matrix $M$ and stiffness matrix $K$. The important computational property is modal decoupling; the two-term formula is a sufficient construction, rather than the only matrix that can possess that property.^[ref]

For constant symmetric $M,K$, with $M$ positive definite, choose a complete set of mass-normalized normal modes as the columns of $\Phi$:Substitute $u=\Phi q$ into $M\ddot{u}+C\dot{u}+Ku=f$ and multiply by ${\Phi }^T$. The result isThis modal damping matrix is diagonal precisely when the chosen modes uncouple the damping. For Rayleigh damping, $D=\alpha I+\beta {\Omega }^2$ is automatically diagonal, and $D_{ii}=2{\zeta }_i{\omega }_i$, where ${\zeta }_i$ is the modal damping ratio. More generally, prescribed diagonal $D$ can be realized by $C=M\Phi D{\Phi }^{T}M$; it need not fit a two-parameter Rayleigh damping curve. OpenSees's separate modal-damping command, for example, accepts individual damping ratios rather than requiring one common pair of mass/stiffness coefficients.^[ref]

For a two-coordinate illustration in consistent normalized units, takeThe undamped mode vectors $(1,1)/\sqrt{2}$ and $(1,-1)/\sqrt{2}$ have natural frequencies $1$ and $\sqrt{3}$. The transformed damping matrix is $D=\mathrm{diag}(0.2,0.4)$, giving damping ratios $0.10$ and $0.2/\sqrt{3}\approx 0.1155$. Here $C=0.1M+0.1K$ explicitly verifies the proportional construction.

In contrast, replacing $C$ by $\mathrm{diag}(0.2,0.4)$ in the original physical coordinates givesThe damping is still dissipative, but the modal equations are coupled: positive dissipated power alone does not imply modal decoupling. Such nonclassical damping can arise when dissipation is distributed differently from the inertial and elastic responses. Dropping its off-diagonal modal terms is an additional approximation that needs justification for the particular system.

Rayleigh damping approximates the damping matrix of a linear structural dynamics model by a combination of its mass matrix and stiffness matrix:Here $\alpha$ has units ${\mathrm{s}}^{-1}$ and $\beta$ has units $\mathrm{s}$. This is a convenient form of proportional damping: for constant symmetric $M,K$ and $M$ positive definite, the same normal modes that separate the undamped system also separate this damping matrix. The OpenSees implementation additionally permits coefficients multiplying current, initial and last-committed stiffness matrices; these are distinct choices in a nonlinear calculation.^[ref]

To see the modal consequence, take an undamped normal mode $K{\varphi }_i={\omega }_i^{2}M{\varphi }_i$ and its modal mass $m_i={\varphi }_i^{T}M{\varphi }_i$. ThenThus Rayleigh damping specifies a frequency-dependent damping ratio, rather than the same damping ratio for every normal mode. The mass-proportional term dominates at low nonzero angular frequency, and the stiffness-proportional term dominates at high angular frequency.

For example, fitting a damping ratio of $0.05$ at both ${\omega }_1=10rad/s$ and ${\omega }_2=50rad/s$ givesThe resulting damping ratio is $0.0375$ at $20rad/s$ and $0.0875$ at $100rad/s$. Agreement at two fitted frequencies therefore does not validate damping outside that band. In particular, a zero-frequency rigid motion has no modal damping ratio of this form, yet the term $\alpha M\dot{u}$ can still resist that motion.

For nonnegative $\alpha ,\beta$ and positive semidefinite $M,K$, the dissipated power ${\dot{u}}^{T}C\dot{u}$ is nonnegative. This supplies an immediate physical check on a fit. It does not establish that the chosen damping matrix represents a particular material or joint: an apparatus would need measured modal decay or another measured dissipative response to justify its coefficients. OpenSees specifically cautions that Rayleigh damping can give incorrect results in nonlinear time-history analysis with concentrated plasticity.^[ref]

Structural dynamics studies the time-dependent motion of structures under applied forces and their response after a disturbance. Inertia, stiffness and damping jointly determine that motion. A common linear small-displacement model iswhere $u$ collects displacement coordinates, $M$ is the mass matrix, $C$ the damping matrix, $K$ the stiffness matrix and $f(t)$ the applied-force vector. Boundary conditions and the chosen coordinates are part of the model; a static force balance omits the velocity and acceleration terms.

A scalar version makes the consequences quantitative. Choose $m=1000\mathrm{kg}$, $k=40000\mathrm{N}{\mathrm{m}}^{-1}$ and $c=400\mathrm{N}\mathrm{s}{\mathrm{m}}^{-1}$. Its undamped natural frequency isFor a constant 100-newton force, static displacement is $100/k=2.5\mathrm{mm}$. For a sinusoidal force $100\cos ({\omega }_{0}t)$, the steady-state displacement amplitude of this linear forced harmonic oscillator is insteadThese are original model values, not measurements of a building. They show why the force amplitude alone does not determine displacement: forcing frequency and damping matter.

Real structures have multiple normal modes. A 2025 study of a viscous inerter damper uses a numerical 20-storey building example and shows that inertia and flexibility associated with discarded modes affect the reduced model's tuning. Matching one nominal mode while ignoring those contributions can therefore misrepresent the coupled device–structure behavior.^[ref]

Measured motion can test a chosen model, while its predictions remain limited by assumptions about linearity, boundary conditions and loading. A large predicted response is a reason to examine those assumptions, not proof that an actual structure will follow the linear response indefinitely.