GOLD STANDARD · POLYMER SCIENCE · HOW X WORKS
Find the decision hidden inside the model
A dielectric loss peak can be broad and skewed at the same time. The Havriliak–Negami response describes both features with two exponents, but those extra degrees of freedom are useful only when the frequency window can identify them.
This guide connects the Gold Standard of Polymer Science, Gold Standard of Education, and How X Works hub.
Chapter contents and reading routes
Choose the route closest to your present task. Every chapter remains expanded, so the article also reads naturally from start to finish.
Orient the response family
Calculate and read the shape
Test identifiability and artefacts
Transfer the analysis
Use evidence and next routes
CHAPTER 1 OF 12 · ORIENT THE MODEL
1. Broad and skewed are different observations
An ideal Debye relaxation has one characteristic time and a symmetric loss peak on a logarithmic frequency axis. Polymer segmental motion often spans a distribution of environments and times, producing a lower, broader peak. The two sides of the peak may also have different slopes.
A Cole–Cole response captures symmetric broadening. Havriliak–Negami adds a second shape exponent so the response can broaden and skew. That makes it a useful empirical family, but it also creates parameter correlations that must be diagnosed.
CHAPTER 2 OF 12 · ORIENT THE MODEL
2. Write the convention before fitting
Using ε* = ε′ − iε″, write ε* − ε∞ = Δε/[1 + (iωτ)^α]^β, with Δε = εs − ε∞ and commonly 0 < α ≤ 1 and 0 < β ≤ 1. Here ω is angular frequency and τ is a characteristic time parameter.
Conventions vary across software and publications. Some use a parameter written 1−α where this guide uses α, or plot positive loss upward despite the minus sign in ε*. State the equation, sign and bounds so fitted numbers can be compared correctly.
CHAPTER 3 OF 12 · READ THE RESPONSE
3. See the nested model family
Setting α = 1 and β = 1 gives Debye. Setting β = 1 gives the Cole–Cole form under this parameter convention, with symmetric broadening controlled by α. Setting α = 1 gives Cole–Davidson, whose β creates an asymmetric response.
The nesting supports an evidence ladder. Start with the simpler model, inspect systematic residuals, and add the second exponent only when the data resolve a real asymmetry. A smaller residual sum alone is not proof of a better physical account.
CHAPTER 4 OF 12 · READ THE RESPONSE
4. Calculate one centre-scale response
Set the dimensionless frequency ωτ = 1, α = 0.8 and β = 0.6. Then (iωτ)^α = exp(i0.4π), and the denominator [1 + exp(i0.4π)]^0.6 can be evaluated directly.
The normalised response (ε*−ε∞)/Δε is approximately 0.697 − 0.276i. Thus ε′ sits about 69.7% of the way from ε∞ to εs, while the positive normalised loss is about 0.276. Unlike Debye at ωτ = 1, the real part is not fixed at one half because the response is asymmetric.
| Model at ωτ = 1 | α | β | Normalised ε′ | Normalised ε″ |
|---|---|---|---|---|
| Debye | 1.0 | 1.0 | 0.500 | 0.500 |
| Cole–Cole example | 0.8 | 1.0 | 0.500 | 0.363 |
| Havriliak–Negami example | 0.8 | 0.6 | 0.697 | 0.276 |
CHAPTER 5 OF 12 · READ THE RESPONSE
5. Separate shape parameters from peak location
The parameter τ sets a characteristic scale, but for a general Havriliak–Negami response the loss maximum is not simply at ωτ = 1. Its location depends on α and β. Treating 1/τ as the measured peak frequency can therefore bias activation or shift-factor calculations.
Use the fitted model to compute the peak location under the stated convention, or report τ as a model parameter rather than relabelling it. Preserve the observed peak and the fitted characteristic time separately.
CHAPTER 6 OF 12 · TEST IDENTIFIABILITY
6. Identifiability needs both flanks and both plateaus
εs, ε∞, Δε, τ, α and β can trade off when the sweep covers only part of the relaxation. A missing low-frequency plateau weakens εs; a missing high-frequency plateau weakens ε∞. If only one flank is measured, τ and skew may compensate for one another.
Profile likelihoods, bootstrap refits or posterior correlations reveal this. Repeat fits after removing an edge decade. If α or β changes sharply while the curve inside the retained window barely changes, the parameter is weakly identified.
CHAPTER 7 OF 12 · TEST IDENTIFIABILITY
7. Do not let conduction become asymmetry
DC conductivity contributes a low-frequency loss that can mimic a long tail. Electrode polarisation, contact resistance, fixture inductance and stray capacitance can distort the same regions that determine β. A flexible response may absorb these artefacts and return plausible-looking exponents.
Fit or remove justified conduction and instrument terms, inspect ε′ and ε″ residuals separately, and repeat thickness, electrode or calibration checks where possible. A parameter is credible when it survives sensible measurement controls.
| Diagnostic | What it tests | Warning sign |
|---|---|---|
| Edge-decade refit | Frequency-window support | Shape exponents move strongly |
| Parameter correlation | Independent information | |correlation| near one |
| Thickness/electrode repeat | Interfacial artefact | Low-frequency shape changes |
| Simpler nested fit | Need for β | No structured residual improvement |
CHAPTER 8 OF 12 · TRANSFER AND AUDIT
8. Transfer from dielectric to mechanical spectra
The same response family is often adapted to complex modulus, compliance or susceptibility. The stable structure is a complex response with a characteristic scale and broadened, possibly asymmetric frequency dependence.
The changed-context check is to rebuild the constitutive meaning. Storage and loss modulus are not numerically interchangeable with permittivity, and transforming between modulus and compliance can change visual peak shape. Refit the correct response quantity rather than copying dielectric exponents.
CHAPTER 9 OF 12 · TRANSFER AND AUDIT
9. Connect temperature without forcing invariance
Temperature shifts the relaxation window and may change broadening. Time–temperature superposition can extend coverage when the same relaxation mechanism and shape persist under horizontal shifting.
If α or β changes systematically with temperature, one master curve may be hiding changing morphology, crystallinity, water content or overlapping transitions. Compare shifted residuals and parameter uncertainty before treating shape as temperature invariant.
CHAPTER 10 OF 12 · TRANSFER AND AUDIT
10. Keep the model empirical and the mechanism testable
The Havriliak–Negami expression was proposed to represent observed polymer dispersions. A good fit describes a response family; it does not uniquely identify a molecular distribution, cooperative length or energy-barrier law. Many microscopic pictures can produce similar spectra.
Use independent evidence such as calorimetry, dynamic mechanical analysis, morphology, composition and ageing response to test a mechanism. Report specimen history, temperature control, frequency window, calibration, equation and uncertainty with the fitted values.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers often ask
Does a β below one prove asymmetric molecular heterogeneity?
No. It shows asymmetry within the chosen empirical model. Overlapping relaxations, conductivity, electrode effects and an incomplete frequency window can produce the same fitted direction.
Should the most flexible model always be preferred?
No. Prefer the simplest model whose residuals and controls support the added shape. If β is unstable or strongly correlated with τ and α, the extra precision in the curve may not be information about the material.
CHAPTER 12 OF 12 · CHECK AND CONTINUE
12. Evidence, connected routes and the next check
Havriliak and Negami’s 1966 polymer-dispersion paper introduces the empirical complex response and shows how it represents Debye, circular-arc and skewed-semicircle behaviour.
Continue through Cole–Cole plots and non-Debye arcs, time–temperature superposition and WLF limits, and shear cessation and structural recovery. The next evidence check is to refit after removing each edge decade and preserve the full parameter-correlation matrix.
Contents · Previous chapter · Return to the Polymer Science root