Differential TET: Correcting Radiation, Residual-Gas Heat Transfer and Coatings

Every measured transient in a real chamber is a sum: the conduction you want, plus radiation from the sample’s surface, plus whatever heat the residual gas still carries, plus — for coated samples — a metallic film’s parallel heat path. Single measurements cannot unmix a sum. But the terms of this particular sum scale differently with things an experimenter controls: radiation’s contribution to apparent diffusivity grows with the square of sample length, a coating’s thermal contribution tracks its electrical conductance — the inverse of resistance. Differential TET is the discipline of measuring along one of those control axes and letting a linear extrapolation separate what a single reading never could.

The transient is a sum

The ideal TET model assumes heat leaves a suspended sample only by conduction to its two heat-sunk ends¹. Reality adds channels — and the smaller the sample, the louder they get: as the field’s canonical review emphasizes, surface-mediated and interface-mediated pathways rise in relative importance precisely as characteristic dimensions shrink². Surface radiation removes heat everywhere along the length; in imperfect vacuum, residual-gas heat transfer does the same³ (“convection” serves in this article as shorthand for that channel; the actual regime — continuum convection, gas conduction or free-molecular transfer — depends on pressure and geometry); on coated nonconductive samples, the metallic film conducts a parallel current of heat alongside the substrate. The distributed loss channels — radiation and residual-gas transfer — accelerate the approach to steady state, and a conduction-only reduction then returns an apparent diffusivity biased upward. A metallic coating is different in kind: it changes both the parallel thermal conductance and the composite heat capacity, so its effect on the fitted value takes its sign from the coating–substrate composite model rather than from one universal rule.

Good vacuum practice suppresses the residual-gas channel; the companion article on the technique’s setup covers that discipline. Radiation, though, cannot be pumped away, and a functional coating cannot be removed. For those two, subtraction has to be analytical — and the analysis works because each channel carries a distinctive fingerprint in how it scales⁴.

The length-squared lever on radiation

The radiation channel’s fingerprint is geometric. For a suspended sample of length L, linearized surface losses contribute to the apparent diffusivity a term proportional to L²·P/A — with P the heated perimeter and A the cross-section — because the loss acts through the surface while conduction escapes through the section, and the conduction time itself scales as L²⁴. For a round fiber P/A reduces the term to the familiar L²/D; for a thin ribbon or film it becomes L²/δ with δ the thickness. The intrinsic diffusivity, by contrast, does not care how long a piece you cut.

That asymmetry is the whole method. Prepare the same material at several suspended lengths; measure the apparent diffusivity of each with the standard reduction⁵; plot apparent α against L². Under the small-rise, constant-property model the points are expected to be approximately linear: the fit’s intercept at L² = 0 is the intrinsic diffusivity — the property of a hypothetical sample too short to radiate meaningfully — and the parasitic channel has been extrapolated away rather than modeled away⁴. No emissivity handbook value was assumed anywhere, which matters because handbook emissivities for processed micro-samples are barely better than guesses.

Interactive: extrapolate against L²

When the slope becomes an effective-emissivity estimate

The extrapolation’s byproduct is as useful as its intercept. Under sufficiently high vacuum — residual-gas transfer negligible or independently corrected — and with the geometry and volumetric heat capacity known, the radiation-dominated slope of apparent α against L² can be inverted for an effective surface emissivity of this specific sample’s surface, in its actual finish and contamination state, at the measurement temperature⁴. If residual-gas transfer is not negligible, the same slope represents a combined distributed-loss coefficient, not emissivity alone — the two channels share the same L²-type geometric signature, which is exactly why the vacuum condition must be established first, not assumed. For micro/nanoscale fibers, whose radiative behavior can depart far from bulk-surface tables, a measured emissivity is data that essentially cannot be bought any other way at this simplicity level — one campaign, two properties.

Three differential axes, one grammar

Reading the three rows together reveals the shared grammar: identify the parasite’s control knob, hold everything else fixed, span the knob honestly, and read the physics at the intercept while the slope pays a dividend. Each axis also cross-checks the others — a length series run at two heating powers, or a power series at two lengths, is a simple two-factor cross-check in which any inconsistency flags a violated assumption before it can contaminate a reported value.

A differential habit, not a single trick

Length-series subtraction is one member of a family-wide habit of mind: whenever a parasite scales differently from the property, measure along the scaling axis and let regression do the separation. The habit predates this particular protocol. The optical-heating/electrical-sensing OHETS technique confronted the same radiative environment in its earliest form, treating conductive and nonconductive micro/nanoscale wires under harmonic optical heating⁷; the steady-state electro-Raman-thermal variant reads a different observable over the same suspended geometry, with its own length-dependent sensitivities⁸. Across all of them the lesson repeats: geometry knobs — length above all — are not nuisance parameters but analytical levers, and the platform-level integration of the family productizes exactly this multi-configuration workflow⁹.

It is worth pausing on what the differential habit buys epistemically. A correction computed from a handbook coefficient inherits the handbook’s error invisibly; a correction extrapolated from the sample’s own measured scaling carries its uncertainty visibly, in the fit statistics. The international uncertainty framework formalizes the distinction — statistically evaluated components versus imported ones¹⁰ — and differential design systematically migrates the radiation term from the imported column to the measured one. Cross-method round-robin experience on microscale wires shows how much of inter-laboratory spread lives in exactly such imported corrections¹¹; a length series is the local antidote.

Instructive contrast comes from methods that handle radiation by construction rather than by extrapolation. Flash analysis operates on disk-shaped bulk samples whose transient is fast and whose standards prescribe explicit radiative heat-loss corrections built into the reduction models¹² — the parasite is modeled, with model-form risk accepted in exchange for single-sample convenience. The pulsed-laser thermal relaxation variant within our own family reads a free cooling decay in which radiative loss is part of the signal being fit rather than a contaminant of it¹³. Neither choice is wrong; they price the same physics differently. The length-differential route’s distinctive offer is that the correction is measured on the specimen itself, with its uncertainty visible in fit statistics rather than buried in a model’s assumptions.

The same logic for coatings

Nonconductive samples enter electrothermal measurement through a thin metallic coating that makes Joule heating and resistive sensing possible — and that coating conducts heat in parallel with the sample it serves. Here the differential axis is not length but the coating itself: its heat conduction ties to its electrical conduction through the Wiedemann–Franz relation, so the coating’s thermal contribution tracks its electrical conductance — the inverse of the measured resistance. Measuring the apparent response across systematically varied coating resistance and extrapolating — conceptually, apparent α against 1/R — separates the substrate’s intrinsic transport from the film’s parallel channel⁴³.

And once again the fit’s parameters carry a bonus: worked in reverse, the analysis yields the effective Lorenz number of the nanoscale metallic film itself. That is not a curiosity. Nanometer-thick films are precisely where the textbook Lorenz value stops being trustworthy: the α_eff–R⁻¹ analysis on iridium films from 7 nm down to sub-nanometer thickness found electrical and thermal conduction both far below bulk, with an effective Lorenz number well above the Sommerfeld value⁶ — so a protocol that measures the film’s own effective value, rather than importing a bulk constant, closes the loop that makes the coating subtraction in the next article of this series defensible.

Designing the length series

A length series succeeds or fails at the sample-preparation bench, before any electronics power up. How many lengths: two points define a line but cannot audit it; use at least three to estimate residual scatter, and preferably four or more when curvature or outliers must be assessed — the count a campaign actually needs follows from the target intercept uncertainty and the mounting realities, not from a fixed rule. Keep the geometry ratio honest: the loss term scales as L²·P/A, so if perimeter or cross-section varies along the series, regress against L²·P/A rather than bare L² — only a constant P/A licenses the shortcut. How to span them: because the loss term rides on L², spread the squared lengths broadly rather than clustering near one end; a geometric progression of physical lengths is one convenient way to do that, though the optimum spacing depends on achievable lengths, remounting error and expected pointwise precision. How to cut them: successive segments from a single continuous specimen, measured longest-first where re-mounting allows, keep composition and processing state constant so the fit’s central assumption — that only length varies — is true by construction rather than by hope. What to log: diameter at each stage (the term scales as L²/D, so an unnoticed taper masquerades as emissivity), mounting torque and paste geometry (contact resistance must not trend with the series), and chamber pressure (a drifting vacuum feeds a second, unmodeled channel into the slope).

The reward for this discipline is a plot whose residuals are as informative as its parameters: flat, structureless scatter supports the selected linear model within the experiment’s sensitivity over the spanned range, while a bow or a fan points at a specific culprit — too-large temperature rises, emissivity drifting with the series, or gas conduction refusing to stay negligible. Run jointly with the zero-rise power series, the two extrapolations bracket the two dominant systematics of suspended-sample work from independent directions — a compact two-factor cross-check that leaves remarkably little room for silent bias.

Discipline and failure modes

Differential methods buy rigor with multiplicity, and multiplicity has its own rules. Sameness across the series: the length series must be the same material in the same state — cut from one specimen where possible — because a differential fit attributes all variation along the axis to the modeled channel; hidden sample-to-sample variation lands directly in the slope and intercept. Span the axis honestly: as with any extrapolation, points clustered at one end make the zero-intercept a long, fragile reach; spread L² over a real range. Per-point quality first: each apparent value should carry its own regression-level precision from a validated transient fit⁵ — the outer line inherits every inner sin. Watch for curvature: a bending trend against L² signals that the linearized radiation treatment is straining (large temperature rises, strongly temperature-dependent emissivity) or a second channel is scaling in; residuals are the diagnostic, and pairing with zero-rise extrapolation keeps the rises small enough for the linear treatment to hold.

These campaigns are routine on our platform: the fiber and film testing service runs length-differential radiation subtraction where sample supply permits, and the 2D material service applies the coating-differential logic that the next article develops in full. The underlying transient model and reduction routes are covered in the TET guide and data-reduction article.

Frequently asked questions

Keep Exploring the ACS Thermal Metrology Knowledge Hub

This article is one chapter of the ACS thermal metrology knowledge hub. To keep going:

• Thermal conductivity & diffusivity testing: the pillar guide — methods, samples and a buyer’s framework in one place.
• The TET technique: a complete guide — how the suspended-sample transient measurement works.
• Zero-temperature-rise extrapolation — removing self-heating bias, validated on a Pt wire.
• The metal-coating protocol — thermal testing for non-conductive samples.
• Fiber & film testing service — wires, fibers and films measured on this platform.

This article presents differential TET analyses — length-series radiation subtraction and coating-differential separation — in general, idealized terms for education. Real campaigns involve material-specific geometry limits, validated linearization ranges and full uncertainty budgets; consult the published protocols and instrument documentation for specifics. The interactive tool above is a schematic teaching aid, not an instrument.