---
title: Dielectric Polarization and Breakdown
module: Capacitance
moduleNumber: 4
lessonNumber: 4
order: 404
summary: |
  Slide a dielectric between the plates and the capacitance rises — but why, and how
  hard can you drive it before the insulator fails? Bound charge answers the first:
  polarization $\vec P$ sets up surface and volume charge that partly cancels the
  applied field, so $\vec D=\varepsilon_0\vec E+\vec P$ separates what the circuit
  controls from what the material contributes. We follow the field across layered
  dielectrics and interfaces, tie permittivity and loss to their frequency dependence,
  and treat dielectric strength as a measured, geometry-dependent limit rather than one
  material number.
topics: [Capacitance]
draft: false
sources:
  - book: Tipler & Mosca
    ref: "Ch. 24 — Capacitance and Dielectrics; §24-5"
---

## Polarization and bound charge

A dielectric contains charges that are bound to atoms, molecules, or a solid lattice
rather than free to cross the sample as they do in a conductor. An applied electric
field can displace positive and negative bound charge slightly, creating induced
dipoles, or can partly orient pre-existing molecular dipoles. The polarization vector
$\vec P$ is the electric dipole moment per unit volume. Its direction is the
direction from bound negative charge toward bound positive charge within the material.

Polarization produces bound charge that reshapes the macroscopic electric field. For
a spatially varying polarization, bound volume charge density is

$$
\rho_b=-\nabla\mathbin{\cdot}\vec P.
$$

At a material surface with outward normal $\hat n$, the bound surface charge
is

$$
\sigma_b=\vec P\mathbin{\cdot}\hat n.
$$

In a uniformly polarized slab, the volume term vanishes in the interior and opposite
bound surface charges appear at the two faces. Their electric field opposes part of
the field produced by free charge on the metal electrodes. Polarization therefore
reduces potential difference for a given electrode charge and increases capacitance
for a given electrode geometry. The bound charges are real charge separations, but
they are not an additional source of freely transferable terminal charge.

$$
% caption: Bound charge from polarization in a dielectric slab. The applied electrode field (grey arrows) separates bound positive and negative charge into aligned dipoles; opposite bound surface charges appear at the two faces, and their own field opposes part of the electrode field inside the material.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[black,fill=black!12] (0.55,2.2) rectangle (5.2,2.42);
\draw[black,fill=black!12] (0.55,0.48) rectangle (5.2,0.7);
\draw[acc,fill=acc!12] (0.55,0.7) rectangle (5.2,2.2);
\foreach \x in {1.0,2.4,3.8} {\draw[->,black,thick] (\x,2.12)--(\x,0.78);}
\foreach \x in {1.7,3.1,4.5} {
  \filldraw[draw=acc,fill=acc!10] (\x,1.62) circle (0.09);
  \filldraw[draw=black,fill=black!25] (\x,1.18) circle (0.09);
  \draw[acc,thick] (\x,1.53)--(\x,1.27);
}
\node[above] at (2.88,2.42) {metal electrode};
\node[below] at (2.88,0.48) {metal electrode};
\node[right] at (5.2,1.45) {polarized slab};
\end{tikzpicture}
$$

The displacement field distinguishes free electrode charge from bound material charge:

$$
\vec D=\epsilon_0\vec E+\vec P,
\qquad
\nabla\mathbin{\cdot}\vec D=\rho_{\rm free}.
$$

Within a specified linear, isotropic range, $\vec P$ is proportional to the
local electric field,

$$
\vec P=\epsilon_0\chi_e\vec E,
\qquad
\vec D=\epsilon\vec E,
\qquad
\epsilon=\epsilon_r\epsilon_0.
$$

The relative permittivity $\epsilon_r=1+\chi_e$ summarizes the linear response.
Its measured value can depend on temperature, frequency, field amplitude, mechanical
stress, and manufacturing history. Quote the test conditions that support the stated
linear approximation.

## Field distribution and microscopic assumptions

A planar dielectric fully filling the gap between large parallel electrodes has its
normal displacement field fixed by free surface charge. With no free charge
inside the slab, the same normal $D$ passes through every interior cross-section,
while the electric field is $E=D/\epsilon$. At the same free electrode charge, the
field is reduced by $\epsilon_r$ relative to vacuum. At a held terminal voltage,
more free charge reaches the electrodes. The two statements use different electrical
boundaries for the same polarization response.

At an interface between two linear dielectrics with no free interfacial charge, the
normal component of $\vec D$ is continuous, whereas normal electric field can
change because permittivity changes. Tangential electric field is continuous in the
electrostatic limit. The boundary conditions distribute field through layered
geometries. A lower-permittivity region carries the stronger local field and can set
the electrical-stress limit even when it occupies a small fraction of the gap.

$$
% caption: Field distribution across two planar dielectric regions. With no free charge at the internal boundary the normal displacement is common to both layers, while the electric-field magnitude differs with permittivity; the lower-permittivity layer (longer arrows) carries the larger local field and can set the usable voltage range.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[black,fill=black!12] (0.55,2.2) rectangle (5.2,2.42);
\draw[black,fill=black!12] (0.55,0.48) rectangle (5.2,0.7);
\draw[acc,fill=acc!12] (0.55,1.45) rectangle (5.2,2.2);
\draw[black,fill=black!8] (0.55,0.7) rectangle (5.2,1.45);
\draw[black,dashed] (0.55,1.45)--(5.2,1.45);
\foreach \x in {1.1,2.0,2.9,3.8,4.7} {\draw[->,acc,thick] (\x,2.0)--(\x,1.6);}
\foreach \x in {1.1,2.0,2.9,3.8,4.7} {\draw[->,black,thick] (\x,1.36)--(\x,0.82);}
\node[left] at (0.55,1.82) {upper};
\node[left] at (0.55,1.06) {lower};
\node[right] at (5.2,1.45) {same D};
\end{tikzpicture}
$$

Microscopic models state the assumptions behind this continuum description. Electronic
polarization follows displacement of electron clouds relative to nuclei and can
respond rapidly. Ionic polarization involves relative displacement of ions in a
lattice. Orientational polarization involves rotation of permanent molecular dipoles
and is slower. Interfacial polarization can accumulate charge near material boundaries
or defects and is often slower still. A measurement at one frequency can sample some
of these mechanisms fully and others only partly, changing both apparent permittivity
and loss. The linear scalar $\epsilon$ model is therefore most reliable for
homogeneous, isotropic material operated well below saturation and away from strong
dispersion bands.

## Dielectric strength and breakdown measurement

Dielectric strength is the electric-field scale at which conduction, damage, or
irreversible change grows rapidly. In an ideal uniform gap, a first estimate is

$$
V_{\rm bd}\approx E_{\rm bd}d.
$$

Actual breakdown follows the largest local field rather than average voltage divided
by nominal thickness. Sharp electrode edges, surface contamination, voids, moisture,
thin spots, and gas pockets concentrate field. A thicker sample can withstand more
voltage while having a lower measured breakdown field because defect statistics and
field uniformity change with thickness. Report dielectric strength with electrode
geometry, thickness, area, voltage ramp rate, temperature, environment, and failure
criterion.

$$
% caption: Breakdown under a controlled voltage ramp. Applied voltage is raised while leakage current is monitored; the reported dielectric strength is tied to electrode geometry, thickness, ramp rate, environment, and a stated failure criterion rather than to voltage alone.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[->,black] (0,0)--(5.4,0) node[right] {applied voltage};
\draw[->,black] (0,0)--(0,3.0) node[above] {leakage current};
\draw[acc,very thick] plot[smooth] coordinates {(0.15,0.08)(1.1,0.12)(2.0,0.18)(2.8,0.32)(3.5,0.85)(4.0,1.9)(4.45,2.7)};
\draw[dashed,black] (3.15,0)--(3.15,0.5);
\node[below] at (3.15,0) {onset};
\node[left] at (4.15,2.45) {rapid conduction};
\end{tikzpicture}
$$

A measurement sequence begins with low-voltage bridge or impedance data to establish
the linear response, followed by guarded leakage measurements at controlled dc
voltage, then a separate breakdown ramp on a fresh specimen. Reusing a sample
after a high-voltage event can change its microscopic state and bias later results.
Current compliance limits damage and provides a defined stop condition. Repeated
specimens are needed because breakdown often has statistical scatter; reporting only
the largest observed voltage gives no estimate of the distribution or the reliability
of the material in the stated geometry.

A quantitative strength estimate must keep voltage and local field separate.

> **Worked example (breakdown voltage estimate).** A uniform gap $d=0.500\ \mathrm{mm}$
> in a material of measured strength $E_{\rm bd}=20\ \mathrm{MV/m}$ withstands, to a
> first estimate,
>
> $$
> V_{\rm bd}\approx E_{\rm bd}d=(20\times10^{6}\ \mathrm{V/m})(0.500\times10^{-3}\ \mathrm m)
> =10.0\ \mathrm{kV}.
> $$
>
> The estimate assumes a uniform nominal gap. A narrow electrode edge or an internal
> void raises the local field above the average by an unknown factor, and long-duration
> operation can reveal leakage or partial discharge below a short-ramp threshold, so
> $V_{\rm bd}$ bounds no particular component's operating voltage on its own.

Design tests therefore include a lower held-voltage
interval with a specified current limit, followed by inspection for irreversible change
and a repeat low-voltage permittivity measurement. A shifted post-test response marks
a changed polarization state even without a visible puncture.

Uncertainty in dielectric strength includes thickness measurement, electrode area and
edge geometry, voltage calibration, ramp rate, temperature, humidity, and the chosen
current or failure threshold. These are not interchangeable random errors: some are
controlled settings that define the reported measurement condition. A complete result
states the weakest observed field, the number of specimens, the distribution summary,
and whether failure was conductive, thermal, or visible damage. Such reporting keeps
microscopic polarization assumptions separate from the empirical limits of a real
dielectric sample.

## Bound charge and the polarization field

Polarization is represented macroscopically by a vector field $\vec P$: dipole
moment per unit volume. Its charge contribution is bound charge. Where polarization
terminates at a material surface, its normal component produces bound surface charge,

$$
\sigma_b=\vec P\cdot\hat n,
$$

and spatial variation within the material produces bound volume charge,

$$
\rho_b=-\nabla\cdot\vec P.
$$

The outward normal in the surface relation belongs to the dielectric body. A uniform
polarization in a rectangular slab has no bound volume charge in its interior, but it
has equal and opposite bound charge on the faces normal to $\vec P$. This is the
macroscopic remnant of many aligned microscopic dipoles: positive ends terminate on
one face and negative ends on the other. The bound charges alter the electric field,
although charge supplied through the external circuit remains the free charge on the
conductors.

The displacement field keeps that bookkeeping explicit:

$$
\vec D=\epsilon_0\vec E+\vec P,
\qquad \nabla\cdot\vec D=\rho_f.
$$

Only free charge appears on the right side of the second relation; the bound-charge
contribution is already contained in $\vec P$. For a linear, isotropic material,
$\vec P=\epsilon_0\chi_e\vec E$, so $\vec D=\epsilon\vec E$ with
$\epsilon=\epsilon_0(1+\chi_e)$. The description applies locally. A single scalar
permittivity requires approximately linear, direction-independent response under the
stated conditions.

## Filled capacitors and interface conditions

In a wide parallel-plate capacitor completely filled by one linear dielectric, the
free surface charge density on the metal fixes the normal displacement field:
$D_n=\sigma_f$. The electric field inside is then $E_n=D_n/\epsilon$, smaller
than the vacuum value at the same free charge. The voltage follows from the line
integral of field, giving $C=\epsilon A/g$ when edge fields are negligible. Bound
charge reduces the field and increases capacitance; the metal-plate area is unchanged.

At an interface between two dielectrics with no free sheet charge, normal displacement
is continuous:

$$
D_{1n}=D_{2n}.
$$

The normal electric field generally changes because $E_n=D_n/\epsilon$. A region
with lower permittivity carries the larger field and hence the larger voltage drop per
unit thickness. Tangential electric field is continuous in electrostatic conditions.
These relations make layered samples a field-distribution problem even before any
breakdown test: the average applied voltage does not identify the field in each layer.

If a free conducting sheet is inserted between dielectric regions, its free surface
charge creates a jump in normal $\vec D$. The boundary condition is
$(\vec D_2-\vec D_1)\cdot\hat n=\sigma_f$. Separate free sheet
charge, bound surface charge, and the field values on the two sides. That bookkeeping
connects a capacitance measurement to a physical interface model.

## Linear-response energy and a measurement geometry

At fixed free charge in a linear dielectric, charging work can be written with the
electric displacement as

$$
u=\int_0^{\vec D}\vec E\cdot \d\vec D'
=\frac12\vec E\cdot\vec D.
$$

In an isotropic medium this becomes $u=\epsilon E^2/2=D^2/(2\epsilon)$. The
one-half again reflects a response built from zero field: both free charge and
polarization grow during charging. This energy expression is a state relation for the
specified linear material and field distribution. A lossy or history-dependent
specimen requires its measured voltage-charge cycle, which can enclose work unrecovered
on discharge.

A guarded measurement separates the central sample response from edge current and
stray capacitance. Place a specimen of measured thickness $g$ between a driven
electrode and a sensing electrode, surround the sensing electrode with a guard held at
the same potential, and keep lead geometry fixed during open and sample readings. A
low-amplitude ac bridge measurement determines capacitance and loss without
approaching the high-field conditions used for leakage or breakdown. Frequency,
temperature, electrode area, thickness, and ac amplitude are part of the result.

Measure the empty fixture, then a reference specimen of known response, then the
sample without changing cables or guards. The reference checks scale and drift; the
empty reading estimates fixture contribution. Report the corrected
capacitance together with the geometry used to infer $\epsilon$. Reliable comparison
requires frequency, temperature, and electrode geometry alongside a reported
“dielectric constant.”

$$
% caption: Guarded dielectric-capacitance measurement. A guard ring at the sensing-electrode potential confines the principal field to the stated sample area, while fixed leads and an empty-fixture reading control the parasitic contribution before permittivity is inferred from geometry.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[black,fill=black!16] (1.15,2.25) rectangle (4.85,2.47);
\draw[black,fill=acc!12] (1.5,0.95) rectangle (4.5,2.25);
\draw[black,fill=black!16] (2.35,0.55) rectangle (3.65,0.77);
\draw[black,fill=black!16] (1.5,0.55) rectangle (2.15,0.77);
\draw[black,fill=black!16] (3.85,0.55) rectangle (4.5,0.77);
\foreach \x in {2.6,3.0,3.4} {\draw[->,acc,thick] (\x,2.05)--(\x,0.9);}
\node[above] at (3.0,2.47) {driven};
\node[below] at (3.0,0.55) {sense};
\node[below] at (1.82,0.55) {guard};
\node[right] at (4.5,1.6) {sample};
\end{tikzpicture}
$$

### Reading a polarization measurement

The capacitance of the guarded cell is a measured terminal response, whereas
polarization is the internal model used to interpret it. In a uniform sample filling
the active area, the low-frequency estimate $\epsilon=Cg/A$ is valid only after
fixture capacitance and lead effects have been treated. A measured increase in
capacitance can arise from the sample, a shifted electrode spacing, a changed guard
connection, or a changed effective area. The geometry must therefore be measured as
carefully as the bridge reading when a permittivity is quoted.

Frequency is especially informative. In a simple linear model, capacitance is nearly
constant and current leads voltage by a quarter cycle. Real samples can show a
frequency-dependent response and a component of current in phase with voltage. The
bridge then reports both a capacitive part and a loss-related part. The loss-related
component records energy dissipation under the stated small-signal conditions. A
frequency sweep should specify the ac amplitude, dc bias if present, settling time,
and whether the reported value came from a series or parallel bridge representation.

Temperature and moisture must be controlled rather than appended as afterthoughts.
They can change molecular mobility, surface conduction, and dimensions of the sample
or fixture. A practical sequence equilibrates the assembled cell at the stated
temperature, records empty and reference readings, mounts the sample without moving
the guarded electrodes, then repeats a low-voltage scan at several frequencies. A
second reading after the sample is removed checks that no contamination or contact
change has altered the fixture baseline.

The measurement can also test the boundary model directly. If the sample is a stack
of two known layers, use the displacement continuity condition to predict the series
voltage division and resulting terminal capacitance. Compare that prediction with the
guarded measurement before assigning an effective permittivity to the stack. An
effective value may be convenient for a particular geometry, but it does not replace
the fields and bound charges in the individual layers. The lower-permittivity layer
can carry the larger electric field even when its contribution to total thickness is
small.

Uncertainty reporting should separate repeatability from calibration and geometry.
Repeated bridge readings estimate short-term scatter. Reference standards and bridge
corrections set capacitance scale. Thickness uncertainty enters directly into
$\epsilon=Cg/A$, and active-area uncertainty enters inversely. Correlations matter:
several readings using one thickness gauge share its scale error, so averaging them
does not remove that contribution. The final record should include the corrected
capacitance, loss quantity, frequency, amplitude, temperature, area, thickness,
guard state, and uncertainty method. These details keep a polarization-field
interpretation tied to an identifiable physical specimen rather than to an isolated
number.

The material inference follows a staged measurement record.

| record | reported quantity | physical inference | configuration held fixed |
| --- | --- | --- | --- |
| empty fixture | terminal baseline and loss | bounds fixture and lead contribution | cables, guard, frequency, amplitude |
| reference specimen | corrected response of known material | checks calibration scale and drift | electrode pressure and active area |
| sample specimen | $\epsilon\simeq C_{\rm corr}g/A$ for a uniform linear region | relates terminal capacitance to the stated geometry | thickness map, guard state, temperature |
| frequency sweep | capacitance and loss versus frequency | resolves dispersive and conductive response | terminal representation and settling time |

The relation for $\epsilon$ applies only after the fixture correction and geometric
model have been justified for the stated cell. An unqualified bridge reading does not
establish a material parameter.

These controls make low-field material data reproducible across laboratories.

Geometry needs its own measurement record. Define the active area by the guarded
electrode overlap; a coupon's outside dimensions can include inactive margin, a guard
gap, and regions with poorly defined pressure. Measure thickness at a grid of
positions across that active region, retain the individual readings, and identify the
reference surfaces used by the thickness gauge. A compressed soft specimen can have a
different thickness under the measurement gauge, inside the test fixture, and after
electrode removal. For thin films on a substrate, state whether the reported
thickness came from a witness sample, a step-height measurement, ellipsometry, or a
cross-sectional measurement, since those methods can sample different locations and
definitions of the film boundary. Edge burrs, particles, trapped air, and tilted
electrodes change local spacing while leaving a nominal thickness unchanged. A
photograph or dimensioned sketch of the guarded stack gives later readers a usable
link between the reported capacitance, the electric-field model, and the physical
sample. Repeat the geometry survey after any fixture change rather than transferring
an earlier correction to a reassembled cell.

Record electrode pressure, clamping sequence, and whether a spacer or compliant
layer fixes separation. These details can shift active area and thickness together,
creating a correlated geometry error that repeated bridge readings cannot reveal.

## Polarization mechanisms and frequency range

Polarization is not one process with one response time. Electronic displacement is
usually able to follow a rapidly changing field, while ionic displacement, molecular
orientation, and motion of larger interfacial charge structures can be progressively
slower. At a frequency well below a mechanism's characteristic rate, that mechanism
contributes to the measured polarization. Above its rate, it cannot follow the field
through a full cycle and its contribution to the in-phase permittivity falls.

The measured response is therefore commonly written with a complex permittivity,

$$
\epsilon^*=\epsilon'-i\epsilon''.
$$

The real part describes the field component stored reversibly in the linear response;
the imaginary part describes a component associated with energy loss per cycle. The
sign convention varies between fields, so a report should state the convention rather
than comparing signs in isolation. A rise or fall in measured capacitance across a
frequency range is meaningful only when electrode polarization, fixture parasitics,
and bridge representation have also been checked.

$$
% caption: Frequency dependence of the dielectric response. Faster polarization mechanisms stay active to higher frequency while slower mechanisms drop out, so the measured permittivity and loss depend on the stated frequency range.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[->,black] (0,0)--(5.4,0) node[right] {frequency};
\draw[->,black] (0,0)--(0,3.0) node[above] {response};
\draw[acc,very thick] plot[smooth] coordinates {(0.25,2.5)(1.1,2.42)(1.9,2.12)(2.65,1.55)(3.55,1.02)(4.7,0.75)};
\node[above] at (1.0,2.42) {slow range};
\node[right] at (3.9,0.95) {fast range};
\end{tikzpicture}
$$

### Loss tangent and dissipated energy

The loss tangent summarizes the ratio of dissipative to stored response,

$$
\tan\delta=\frac{\epsilon''}{\epsilon'}.
$$

In a sinusoidal test at angular frequency $\omega$, a capacitor with loss has a
current component in phase with voltage as well as the quadrature capacitive current.
The in-phase component is responsible for average power dissipation. High frequency
or large voltage can produce appreciable heating even at small loss tangent because
power grows with frequency and field amplitude. At low frequency, dc leakage or
electrode effects can produce a large apparent loss without identifying a bulk
polarization mechanism.

> **Worked example (dielectric heating).** A $C=10\ \mathrm{nF}$ capacitor with loss
> tangent $\tan\delta=0.020$ carries $V_{\rm rms}=100\ \mathrm V$ at $f=1.0\ \mathrm{kHz}$.
> Its equivalent parallel conductance is $G=\omega C\tan\delta$, so the average
> dissipated power is
>
> $$
> P=V_{\rm rms}^2\,\omega C\tan\delta
> =(100)^2(2\pi\cdot1.0\times10^{3})(10\times10^{-9})(0.020)
> =13\ \mathrm{mW}.
> $$
>
> The same tangent at $100\ \mathrm{kHz}$ raises the dissipation a hundredfold, to
> $1.3\ \mathrm W$. At small loss tangent the heating scales as $\omega C V^2$, so
> frequency and amplitude, not the tangent alone, decide whether a component runs warm.

Use a calibrated bridge or impedance analyzer with a stated equivalent-circuit mode.
Series and parallel loss representations can give different numerical resistance
values while describing the same measured admittance near one operating point. Record
the chosen representation, ac amplitude, dc bias, temperature, and electrode state.
Those conditions distinguish a material comparison from an instrument setting.

$$
% caption: Loss-angle interpretation. The capacitive current leads the voltage by a quarter cycle, while the small in-phase component represents energy dissipated each cycle; the loss tangent is the ratio of the in-phase part to the quadrature part.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[->,black] (1.0,0.7)--(4.6,0.7) node[right] {voltage};
\draw[black,dashed] (1.0,0.7)--(1.0,2.95);
\draw[->,acc,very thick] (1.0,0.7)--(2.6,2.7);
\draw[black,dashed] (2.6,2.7)--(1.0,2.7);
\node[left] at (1.0,1.85) {quadrature};
\node[above] at (1.7,2.72) {loss part};
\node[above right] at (2.55,2.68) {current};
\end{tikzpicture}
$$

## Breakdown pathways and guarded test protocol

The observed breakdown threshold depends on defects, voids, sharp electrode features,
thermal hot spots, and conductive surface paths. Thickness matters because a thicker
specimen samples a longer path containing more possible weak regions, while area
matters because a larger electrode area samples more material volume and edge length.
Consequently, nominally identical samples can fail at different fields. Report a
specimen count and distribution summary alongside strength rather than only a maximum
or mean voltage.

Use separate specimens for destructive ramps. First measure low-field capacitance and
loss in the guarded cell. Mount a fresh specimen with clean electrodes, a defined
guard connection, a calibrated thickness reading, and a current-limited high-voltage
source. Raise voltage at a stated rate while logging terminal voltage and leakage
current. Stop at a predeclared current threshold or an irreversible change in the
trace. The guard reduces surface leakage into the sensing path. Electrode finish and
ambient conditions still determine the local field at an imperfect edge and belong in
the test description.

After the ramp, discharge through a rated path before handling the fixture. Inspect
the specimen and repeat the low-field measurement only when the protocol calls for a
post-event comparison. A changed capacitance or loss reading is evidence that the
sample state has changed even if no hole is visible. Pool results only from specimens
prepared and tested under the same thickness, electrode, ramp, temperature, and
humidity conditions.

### Frequency response versus breakdown data

Frequency-response data and breakdown data answer different questions. Small-signal
permittivity describes the reversible and dissipative response under the specified ac
field. A high-voltage ramp probes leakage growth, local heating, and eventual failure
in one particular electrode geometry. A specimen with low loss at one frequency can
still fail early because of a void or an edge defect, while a specimen with measurable
loss can remain stable at a modest field when heat is removed effectively. Do not use
one result as a substitute for the other.

The time scale of the test must be retained. A bridge measurement may use milliseconds
per cycle, a leakage hold may last minutes, and a service exposure may last years.
Slow charge motion and conduction paths can be nearly invisible in a short ac scan but
important during a dc hold. Conversely, a rapid high-voltage ramp can reach a larger
voltage before heat or charge redistribution develops fully. Report ramp rate, hold
time, sampling interval, and current compliance alongside the voltage threshold.

Statistical reporting is particularly important for breakdown. Sort the failure fields
from several fresh specimens and show their spread or fitted probability distribution.
The weakest measured value supports conservative screening, but it is not the same
quantity as a characteristic field from a distribution model. If specimens have
different areas or thicknesses, retain those values individually; pooling them without
geometry information hides the dependence that the test is meant to reveal.

Guarding helps identify unwanted surface paths during low-current measurements, yet a
guard connection must not be allowed to become an uncontrolled third electrode during
the high-voltage ramp. Specify its potential, spacing from the active electrode, and
whether it remains connected throughout the test. Clean, dry handling procedures and
the elapsed time after cleaning are also relevant because surface contamination can
change leakage by orders of magnitude without altering the bulk polarization response.

A dielectric report needs three separate records: low-field capacitance and loss at
stated frequency and temperature, the dc leakage curve at stated field and duration,
and the destructive threshold distribution for fresh specimens. Separate records
preserve the distinct evidence for material response, conductive stability, and
failure reliability.

Instrument limits should be checked against the expected current scale before a test
begins. The current range must resolve the low-field leakage of interest while also
surviving the predeclared compliance limit without losing time resolution. Voltage
measurement should be made at the specimen electrodes whenever lead resistance or
protective elements can create a difference from the source indication. For ac work,
verify that the bridge stays within its phase and amplitude accuracy over the selected
frequency range; a changing fixture baseline can otherwise look like a material
dispersion.

Repeat measurements in a planned order. A low-field frequency sweep before and after
a moderate dc hold can reveal reversible settling without destroying the specimen. A
breakdown ramp belongs last and uses a fresh sample when a distribution is required.
Archive the raw time traces as well as reduced values. The shape of current before the
stop condition can distinguish a sudden conductive event from a gradual thermal or
surface process, which matters when comparing samples with similar reported threshold
fields.

Before comparing specimens, confirm that the same electrode preparation, storage time,
and ambient conditioning were used. Otherwise an apparent material trend may simply
be a difference in surface state. A written test sheet with these controls prevents
that ambiguity when results are revisited later.

Document every deviation from the planned sequence, including interrupted ramps,
instrument range changes, and any visible specimen handling event.

## Layered dielectrics, energy, and leakage

In dielectric layers placed one after another across the field direction, normal
displacement is common to every layer when no free charge lies at an internal
boundary. If layer $j$ has thickness $g_j$, area $A$, and permittivity
$\epsilon_j$, then

$$
V=\sum_j E_jg_j=\frac{Q}{A}\sum_j\frac{g_j}{\epsilon_j},
\qquad
C=\frac{A}{\sum_j g_j/\epsilon_j}.
$$

The lower-permittivity layer has the larger field. Its energy density
$u_j=D^2/(2\epsilon_j)$ is also larger at the shared displacement field, so a thin
low-permittivity layer can dominate both voltage drop and stored energy. Determining
that redistribution requires thickness and interface geometry in addition to terminal
capacitance.

Leakage adds a parallel conduction path to the displacement response. A simple dc
model is $I=V/R_\ell+C\,\d V/\d t$. Under a voltage step, the second term is transient;
the later current estimates a conductance only if absorption and temperature drift are
small. Fit current against hold time rather than selecting one early reading. A stack
can have the same terminal capacitance as another stack yet a very different leakage
path because conduction is controlled by interfaces and the least resistive region.

Characterization uncertainty follows the model. Thickness errors enter each
$g_j/\epsilon_j$ term; area uncertainty is shared; bridge scale and open-fixture
correction affect the terminal capacitance. Report correlated geometry errors rather
than treating every layer measurement as independent. Repeat specimens distinguish
material spread from meter repeatability.

$$
% caption: Leakage-current interpretation. A voltage step drives a fast capacitive transient followed by a slower conduction current; the hold-time record is needed before a leakage resistance is inferred for a layered specimen.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[->,black] (0,0)--(5.3,0) node[right] {time};
\draw[->,black] (0,0)--(0,2.9) node[above] {current};
\draw[acc,very thick] plot[smooth] coordinates {(0.25,2.5)(0.65,1.7)(1.15,1.1)(1.95,0.78)(3.15,0.65)(4.8,0.6)};
\node[above right] at (0.65,1.75) {step transient};
\node[above] at (3.8,0.65) {conduction};
\end{tikzpicture}
$$

Measure empty fixture, reference sample, and specimen with unchanged guards and leads.
Record frequency, voltage, temperature, layer thicknesses, area, and hold-time
current. Propagate bridge, geometry, and calibration terms to the inferred
permittivity or leakage value, then identify terms shared across specimens. This
record supports a layered-field interpretation instead of an unsupported
effective-material number.

### Multilayer voltage division

For a two-layer specimen of common area $A$, thicknesses $g_1,g_2$, and
permittivities $\epsilon_1,\epsilon_2$, free charge $Q$ on the electrodes sets the
displacement $D=Q/A$. The separate fields and voltage drops are

$$
E_1=\frac{D}{\epsilon_1},\quad E_2=\frac{D}{\epsilon_2},\quad
V_1=\frac{Dg_1}{\epsilon_1},\quad V_2=\frac{Dg_2}{\epsilon_2}.
$$

Then $V_1/V_2=(g_1/\epsilon_1)/(g_2/\epsilon_2)$, so a layer can carry most of the
terminal voltage without being the thickest. Compare every $E_j$ with the relevant
low-field and failure data; the average $V/(g_1+g_2)$ omits the interface
distribution. The energy per area is $\sum_j D^2g_j/(2\epsilon_j)$, so the
low-permittivity layer also carries the larger energy density at common displacement.

### Leakage over time and temperature

Following a dc step, current can decline for several reasons: the ideal displacement
current vanishes quickly, slower polarization relaxes, and true conduction approaches
a later value. Plot current against logarithmic time across the full hold interval.
Represent the late current by a conductance only after the curve has approached a
stable range. If it continues to drift, report the hold time and current history
instead of assigning one resistance.

Temperature changes both molecular mobility and conduction pathways. A temperature
sweep must allow the specimen to equilibrate before each reading; otherwise a thermal
transient can resemble dielectric absorption. Keep humidity, electrode pressure, and
guard condition fixed. Comparing current at two temperatures is meaningful only when
the same terminal voltage and specimen coordinate are used.

### Uncertainty and traceability

For the example above, thickness uncertainty enters twice: it changes the denominator
of terminal capacitance and changes the calculated field in each layer. Area scale,
bridge calibration, reference standard, and fixture correction affect capacitance.
The layer permittivities may be measured separately or taken from a traceable reference
under matching frequency and temperature. State which quantities are measured on the
assembled stack and which are imported assumptions.

Repeat a complete sequence on several specimens. Keep raw bridge readings, empty-cell
and reference checks, thickness maps, temperature record, terminal-voltage trace, and
timed leakage record. The archive separates uncertainty in terminal capacitance from
uncertainty in inferred layer fields. It also identifies whether a disagreement arises
from interface geometry, sample variation, or a changed fixture baseline rather than
from an unexplained effective permittivity.

### Two-layer interface conditions

At a planar dielectric interface without free sheet charge, the normal component of
$\vec D$ is continuous. One free electrode charge density sets both layer
displacement fields in a two-layer test cell. Equal electric fields require
equal permittivities:

$$
\epsilon_1E_1=\epsilon_2E_2=D.
$$

The voltage is the sum of the layer drops. Geometry gives
$V=D(g_1/\epsilon_1+g_2/\epsilon_2)$. A physical material stack therefore follows
a series field geometry. The thin or low-permittivity layer can control the highest
local field. Averaging thickness first and assigning one permittivity loses that
information.

> **Worked example (two-layer voltage division).** A test cell of active area
> $A=10.0\ \mathrm{cm^2}$ holds two layers: $g_1=0.200\ \mathrm{mm}$ with
> $\epsilon_{r1}=2.0$, and $g_2=0.300\ \mathrm{mm}$ with $\epsilon_{r2}=6.0$. The
> series stack has capacitance
>
> $$
> C=\frac{\epsilon_0A}{g_1/\epsilon_{r1}+g_2/\epsilon_{r2}}=59.0\ \mathrm{pF}.
> $$
>
> With $100\ \mathrm V$ applied, displacement continuity fixes the voltage ratio
> $V_1:V_2=(g_1/\epsilon_{r1}):(g_2/\epsilon_{r2})=2:1$, so the drops are
> $66.7\ \mathrm V$ and $33.3\ \mathrm V$ and the local fields are
> $0.333\ \mathrm{MV/m}$ and $0.111\ \mathrm{MV/m}$. A guarded bridge reading near
> $59.0\ \mathrm{pF}$ confirms the terminal model but not the layer fields; those
> follow from the boundary condition and the measured thicknesses.

Use the same guarded electrodes for empty-fixture, reference, and stack readings.
Measure thickness at several points, record temperature and frequency, and keep cable
routing unchanged. The uncertainty in $g_1$ affects both terminal capacitance and
the larger calculated field, so it can dominate the uncertainty in the limiting layer.

$$
% caption: Voltage division across a two-layer stack. Displacement continuity fixes the ratio of the layer voltage drops from the thickness-to-permittivity ratios, so a layer can take most of the terminal voltage without being the thickest.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[->,black] (0,0)--(5.2,0) node[right] {layer};
\draw[->,black] (0,0)--(0,2.9) node[above] {voltage drop};
\draw[acc,fill=acc!14] (1.0,0) rectangle (2.2,2.3);
\draw[black,fill=black!10] (3.0,0) rectangle (4.2,1.15);
\node[below] at (1.6,0) {layer one};
\node[below] at (3.6,0) {layer two};
\node[above] at (1.6,2.3) {larger drop};
\node[above] at (3.6,1.15) {smaller drop};
\end{tikzpicture}
$$

### Ageing, humidity, and partial discharge

Dielectric condition can change before a complete puncture. Repeated field exposure
may alter interfaces, create conductive paths, or enlarge microscopic voids. Partial
discharge occurs when a local gas-filled region or defect becomes conductive for part
of an ac cycle while the bulk specimen still supports the applied voltage. Its pulses
can erode nearby surfaces and progressively lower the margin to failure. Void
geometry, pressure, frequency, waveform, and electrode finish determine onset; the
average field alone is insufficient.

Humidity can increase surface leakage and change interfacial chemistry. Temperature
changes conductivity, relaxation time, and mechanical dimensions. Precondition every
specimen by a stated temperature-humidity history, then allow the guarded fixture to
equilibrate before low-field and high-field readings. A result from a dry freshly
baked specimen should not be compared directly with one exposed to laboratory air
without recording that difference.

$$
% caption: Ageing pathway in a dielectric specimen. A local discharge in a gas-filled void or moisture-assisted surface leakage can change a defect region before complete breakdown, so low-field response and leakage history are watched alongside destructive tests.
\begin{tikzpicture}[>=stealth,font=\footnotesize]
\definecolor{acc}{HTML}{4A6FA5}
\draw[black,fill=black!16] (0.7,2.2) rectangle (5.2,2.42);
\draw[black,fill=black!16] (0.7,0.5) rectangle (5.2,0.72);
\fill[acc!12] (0.7,0.72) rectangle (5.2,2.2);
\foreach \x in {1.15,1.7,4.5,4.95} {\draw[->,acc,thick] (\x,2.12)--(\x,0.8);}
\filldraw[draw=black,fill=black!8] (2.8,1.35) circle (0.17);
\draw[acc,very thick] (2.8,1.52)--(2.7,1.78)--(2.9,1.98)--(2.8,2.14);
\node[right] at (2.99,1.3) {void};
\node[right] at (2.95,2.0) {discharge};
\end{tikzpicture}
$$

### Breakdown data and conservative operation

Use fresh specimens for destructive ramps and retain the full voltage-current trace.
The interpretation begins with the test condition: thickness map, active area,
electrode preparation, guard state, temperature, humidity, ramp rate, current limit,
and the rule that ended the test. Sort failure fields from all specimens and report a
distribution or at least median, range, and specimen count. An operating rating
requires more than a single largest value.

Separate a short-ramp threshold from an operating field. Operating service adds time,
thermal cycling, waveform peaks, humidity excursions, and manufacturing variation.
A conservative report states the lowest tested field without irreversible change over
a specified hold time, then applies a documented margin for the intended service
condition. Scatter and ageing evidence inform that engineering margin; no universal
fraction follows from a laboratory breakdown value.

Compare pre-test and post-test low-field capacitance, loss, and leakage on specimens
that have not been destroyed. A drift in these quantities can reveal condition change
before a visible failure. If pulse monitoring is available, record its threshold and
count rate together with the applied waveform. These data distinguish a stable sample
from one that merely survived a short high-voltage ramp.

### Service qualification

A dielectric qualification begins with a specimen population rather than one
exceptional breakdown value. Cut or fabricate specimens with recorded batch, cure
history, thickness map, electrode geometry, and surface preparation. Store them under
the stated humidity and temperature condition long enough for mass and electrical
response to settle. Randomize the order of specimens during a test series so a slow
change in laboratory humidity, electrode wear, or instrument zero does not become a
false material trend. Retain unused reference specimens alongside stressed specimens
when the test programme includes long holds or repeated cycling.

Separate destructive ramps from nondestructive qualification. A ramp finds a failure
distribution under one waveform, ramp rate, and current limit. A service test holds a
lower field for a declared duration while recording leakage, temperature, partial
discharge count if available, and changes in capacitance or loss. It tests whether a
specimen maintains the specified response in one operating envelope. A sample can
survive a fast ramp yet show increasing leakage or discharge activity during a long
hold. A sample can also fail a rapid ramp at a field well above any intended operating
level while remaining suitable for lower service stress. State the conclusion
supported by each dataset.

Field calculations need the smallest plausible local gap and the highest plausible
local voltage. Thickness variation, a particle under an electrode, a sharp conductor
edge, and a void inside a laminate all raise local stress relative to the nominal
applied-voltage divided by average thickness. Map thickness over the active area,
inspect electrode finish, and document any guard or edge-radius treatment. When
geometry uncertainty prevents a credible local-field estimate, report terminal
voltage and specimen geometry separately instead of claiming an overly precise
breakdown field.

Ageing tests should define the cycling waveform and rest interval. Repeated dc
polarity reversals, ac cycles, temperature steps, and humidity exposures can activate
different transport and mechanical processes. Measure a low-field reference response
before the series, between exposure blocks, and after recovery time. A monotonic
increase in leakage, a shift in loss peak, or a rising pulse count gives an earlier
warning than final puncture. A reversible temperature response can be separated from
permanent degradation by returning the conditioned specimen to its reference
temperature and repeating the low-field measurement after a specified dwell.

Uncertainty in a service margin combines material scatter and test definition. Do not
average failure fields from different thickness ranges, waveforms, or humidity states
into one undifferentiated number. Group records by matched condition, then report
specimen count, central tendency, spread, censoring rule for unfailed holds, and the
chosen margin. An operating field derived from the lowest qualified hold condition is
conservative only for conditions that match the tested geometry, waveform, thermal
path, and environment. Changes in any of those conditions require a new qualification
or an explicitly justified derating.

The final data package contains raw voltage-current records, current-limit events,
timed leakage traces, temperature and humidity logs, specimen photographs, thickness
maps, calibration identifiers, and a table linking each specimen to its preparation
and outcome. It supports later comparison with a revised electrode design or material
batch and ties each service limit to its supporting evidence.

A qualified field applies only within its tested thickness and waveform envelope until
local-field analysis is repeated for a new design. A thinner layer can change voltage
division, heat removal, defect statistics, and the relevance of a surface treatment.
A new electrode shape can move limiting stress from the bulk to an edge or interface.
State the tested envelope and calculated local-field assumptions beside every operating
limit. This ties a conservative service rating to the material system that was
measured.

Archive the rejected records as well as the successful holds. A failed guard check,
an unstable leakage trace, or a specimen damaged during mounting identifies the
practical boundary of the measurement procedure. Future qualification work can then
separate material variation from a changed fixture, conditioning history, or
analysis rule.