The Gold Standard of X-ray Photoelectron Spectroscopy | Why Two Angle-Resolved Signals Do Not Uniquely Determine a Depth Profile

A concentration depth profile describes how an analyte fraction changes with distance below a surface. The instrument does not report that entire function directly. It reports photoelectron signals influenced by emission, transport through material and detection. Recovering the profile from those signals is an inverse problem.

Before fitting, state the unknown. Are we estimating one uniform film thickness, three layer concentrations, or an unrestricted continuous function? Those are different problems even when the same instrument supplies the data. A model that identifies one thickness parameter need not identify an arbitrary depth distribution.

Here the layer boundaries are independently known and only three concentrations are unknown. This deliberately simple case makes the information boundary visible with ordinary algebra. If nonuniqueness already appears here, a more flexible profile requires at least as careful an account of what the measurements constrain.

Contents · Next chapter

Let z be depth measured inward from a flat surface, with z = 0 at the surface. Let θ be the emission angle measured from the outward surface normal. At normal emission θ = 0. At a more grazing direction θ increases, while cos θ decreases.

An electron emitted at depth z follows a straight path of length z/cos θ in the simplified transport picture. If λ is a positive attenuation length, the survival weighting is exp[−z/(λ cos θ)]. Deeper contributions are reduced more strongly at larger θ.

This convention must be explicit. Some instruments or texts specify an angle relative to the surface plane instead. Copying a numerical angle without converting its convention can reverse the intended geometry. Our examples consistently use the surface normal, and every angle-dependent factor follows that definition.

Contents · Previous chapter · Next chapter

Assume an analyte fraction c(z) between zero and one, a known constant atomic density, a known angle-dependent response factor Kθ and a single known λ. Write Iθ = Kθ ∫ c(z) exp(−qθz) dz, where qθ = 1/(λ cos θ), and the integral runs over depth from zero to infinity.

Kθ contains the response factors required to turn the concentration integral into the measured intensity. We assume those factors have been independently calibrated for the chosen signal and geometry. An unknown response factor would introduce another unknown into the inverse problem.

The model also stipulates that λ is the same in all the layers. That is a teaching simplification, not a claim that attenuation is composition-independent in every material. We exclude roughness, lateral variation, uncorrected elastic scattering and depth-dependent density. Later chapters return to those limits rather than silently treating them as solved.

Contents · Previous chapter · Next chapter

Let each slab have thickness d. The first occupies 0 ≤ z < d, the second d ≤ z < 2d and the third 2d ≤ z < 3d. Their constant analyte fractions are c0, c1 and c2. Below 3d the analyte fraction is independently known to be zero.

The zero baseline matters. If the deeper substrate also contained an unknown analyte contribution, we would need another term and another unknown. The finite three-slab problem is not a universal description of an XPS sample; it is a precisely bounded reconstruction exercise.

The concentration vector is therefore (c0,c1,c2), with each component between zero and one. The layer boundaries and attenuation model are supplied independently. Our goal is to determine what two angle-resolved signals say about those three fractions, without substituting a preferred shape for an observed one.

Contents · Previous chapter · Next chapter

Define rθ = exp(−qθd), which lies between zero and one. The first slab contributes Kθc0(1−rθ)/qθ. The second contributes the same factor multiplied by c1rθ, and the third contributes it multiplied by c2rθ².

The full signal is therefore Iθ = Kθ(1−rθ)(c0+c1rθ+c2rθ²)/qθ. The weights are integrated attenuation weights, not values of the exponential sampled at three arbitrarily chosen depths.

For convenient algebra define Sθ = qθIθ/[Kθ(1−rθ)]. Then Sθ = c0+c1rθ+c2rθ². This S is a calibrated transformed signal. It is not a raw intensity, an atomic percentage or a probability. Dividing by the angle-dependent factor is legitimate here because qθ, Kθ and d are stipulated known. Without that calibration, the polynomial equation would not describe the recorded numbers directly.

Contents · Previous chapter · Next chapter

Set d = λ ln 2. At normal emission, q0d = ln 2, so r0 = 1/2. At θ = 60°, cos θ = 1/2 and q60 = 2q0. Hence q60d = 2 ln 2 and r60 = 1/4.

Our two equations become S0 = c0+c1/2+c2/4 and S60 = c0+c1/4+c2/16. The more grazing signal suppresses the deeper slabs more strongly, exactly as expected. It still contains contributions from all three slabs.

The values d and λ need not be assigned arbitrary nanometre numbers to demonstrate the inverse structure. Their stated ratio is what fixes the weights. If a real experiment supplies different thicknesses or attenuation lengths, calculate its own weights rather than forcing these convenient fractions onto it. The rational numbers make this teaching example easy to verify independently.

Contents · Previous chapter · Next chapter

Take a uniform three-slab analyte fraction c = (0.4,0.4,0.4), followed by the known analyte-free baseline. The transformed normal signal is 0.4(1+1/2+1/4) = 0.7. The transformed 60° signal is 0.4(1+1/4+1/16) = 0.525.

These two readings are noiseless synthetic values generated by our forward equation. They are deliberately favourable to reconstruction: the angle, thickness, attenuation length and response calibration are all known exactly. Any ambiguity we find is therefore not caused by noisy instruments or poorly estimated response factors.

The uniform profile is one valid explanation of the two signals. The next question is whether it is the only valid explanation. A fit that recovers it has not answered that question until alternative physical profiles have been tested.

Contents · Previous chapter · Next chapter

Now choose c = (0.45,0.1,0.8). Every fraction remains between zero and one, but the depth distribution is markedly different. Its normal signal is 0.45+0.05+0.2 = 0.7. Its 60° signal is 0.45+0.025+0.05 = 0.525.

Both transformed signals agree exactly with those of the uniform profile. Because the known normalisation factors are identical for the two profiles at each angle, their raw intensities also agree at each angle. This is not an ambiguity created by plotting rescaled data.

The second profile has a shallow concentration increase, a depleted middle slab and a rich deeper slab. Their weighted contributions compensate in both measurements. A method that reports the uniform profile uniquely must therefore be using some information or restriction beyond these two ideal readings and the physical fraction bounds.

Contents · Previous chapter · Next chapter

A concentration change Δc leaves both signals unchanged if Δc0+Δc1/2+Δc2/4 = 0 and Δc0+Δc1/4+Δc2/16 = 0. Subtracting the equations gives Δc1 = −3Δc2/4. Substituting back gives Δc0 = Δc2/8.

Thus the vector n = (1/8,−3/4,1) is invisible to both measurements. Every profile c(ε) = (0.4,0.4,0.4)+εn produces the same two signals wherever its fractions remain physical. The alternative profile above corresponds to ε = 0.4.

The same result can be seen as a polynomial. The concentration change contributes ε[r²−3r/4+1/8] = ε(r−1/2)(r−1/4). It vanishes at the two measured r values. This factorisation makes the cancellation exact and shows why a new angle may reveal it.

Contents · Previous chapter · Next chapter

Physical fractions require 0 ≤ 0.4+ε/8 ≤ 1, 0 ≤ 0.4−3ε/4 ≤ 1 and 0 ≤ 0.4+ε ≤ 1. Intersecting these three restrictions gives −0.4 ≤ ε ≤ 8/15, approximately 0.53333.

The fraction bounds therefore remove many mathematical solutions but leave an entire interval of valid profiles. At ε = −0.4 the profile is (0.35,0.7,0). At ε = 8/15 it is (7/15,0,14/15). Both endpoints give exactly the original two signals.

This is a useful distinction when interpreting constrained fitting. Enforcing nonnegative concentrations is physically valuable, but it does not automatically make the answer unique. To claim uniqueness, examine the feasible set after all supplied constraints have been applied. Here that examination can be done explicitly rather than inferred from an optimiser returning one result.

Contents · Previous chapter · Next chapter

Across the feasible family, c0 ranges from 0.35 to 7/15, c1 from zero to 0.7 and c2 from zero to 14/15. These individual ranges are correct, but they cannot be combined arbitrarily. All three concentrations depend on the same ε.

For example, the largest c0 and largest c2 occur together with c1 = 0. The largest c1 occurs at the opposite endpoint, with c2 = 0. Choosing all three individual maxima would not reproduce the measurements and would not belong to the feasible family.

A table of independent-looking error bars can therefore conceal a strong relationship between inferred layer values. Report the family equation or its parameter correlation when that structure matters to the decision. The physically allowed uncertainty is a connected set of profiles, not a box of freely interchangeable layer concentrations.

Contents · Previous chapter · Next chapter

With constant atomic density and equal slab thicknesses, analyte amount per unit area is proportional to d(c0+c1+c2). The uniform profile has concentration integral 1.2d. The alternative profile has integral 1.35d. They produce the same two measured signals despite containing different total analyte amounts in the stipulated depth interval.

For the full feasible family, c0+c1+c2 = 1.2+3ε/8. Its range is 1.05 to 1.4. Multiplying by d and the independently known atomic density converts that concentration integral into an areal atom count within the model.

This provides a second inference limit beyond the layer-by-layer ambiguity. Two attenuated signals do not necessarily determine even the unweighted total amount over a known support. The exponential weights are not uniform weights, and a measured weighted sum should not be silently relabelled as a total inventory.

Contents · Previous chapter · Next chapter

Define Jθ = qθIθ/Kθ = (1−rθ)Sθ. For the original readings, J0 = 0.35 and J60 = 0.39375. Their ordering differs from that of S0 and S60 because the transformation includes an angle-dependent factor.

If K0 = K60 in this stipulated example, the actual raw intensity ratio is I60/I0 = (0.39375/0.35)(q0/q60) = 0.5625. Thus the raw 60° signal is smaller even though J60 exceeds J0. There is no physical contradiction: these are differently defined quantities.

This example is a practical safeguard against reading a processed graph without its normalisation. Record whether a plotted angular series represents raw area, response-corrected area, an atomic fraction or a transformed integral. The algebra must use the quantity actually supplied, including every angle-dependent response and geometric factor.

Contents · Previous chapter · Next chapter

Choose a third angle with cos θ3 = 1/3, so θ3 is approximately 70.53° from the surface normal. With the same d = λ ln 2, q3d = 3 ln 2 and r3 = 1/8.

The uniform profile predicts S3 = 0.4(1+1/8+1/64) = 0.45625. The alternative predicts 0.45+0.1/8+0.8/64 = 0.475. The difference is 0.01875. The two previously indistinguishable profiles now make different predictions.

The family factorisation explains the result: ΔS3 = ε(1/8−1/2)(1/8−1/4) = 3ε/64. At ε = 0.4, this gives 0.01875. The new angle helps because it adds a weighting that does not annihilate the old invisible direction. Repeating one of the original angles more precisely would not do that.

Contents · Previous chapter · Next chapter

At three distinct r values, the response matrix has rows (1,r,r²). Its determinant is nonzero because the r values are distinct. Consequently three exact transformed signals identify the three slab concentrations under the fully specified model.

This is a conditional uniqueness result. It assumes the layer boundaries, known zero baseline, attenuation length and response factors are correct. Allow four unknown slabs or an unrestricted continuous profile and the parameter space changes. Three measurements do not magically identify every possible depth function.

A useful report states both the recovered values and the model space in which they are unique. “Unique three-slab solution with supplied boundaries and transport parameters” is a more precise claim than “the depth profile was directly measured.” The distinction matters when a scientific decision depends on an unmodelled narrow layer or a deeper contribution.

Contents · Previous chapter · Next chapter

Uniqueness does not guarantee high precision. Suppose the first two signals are held exact and the third transformed signal has uncertainty ±0.01. Because ΔS3 = 3ε/64, the corresponding ε uncertainty is ±0.01 × 64/3, approximately ±0.21333, before applying the physical bounds.

That is a substantial concentration-family uncertainty for a comparatively small signal uncertainty. The third angle distinguishes the profiles in principle, yet its sensitivity to this particular direction is limited. If all three readings have uncertainty, propagate their full joint uncertainty rather than treating the first two as exact.

Angles that are very similar produce similar weightings and can make reconstruction particularly sensitive. More separated angles may help the algebra but can also require closer attention to the transport and geometry assumptions. Designing useful measurements balances independent sensitivity with a forward model that remains credible in the chosen configuration.

Contents · Previous chapter · Next chapter

Consider fitting the two original signals while penalising adjacent concentration differences. A simple objective is the sum of squared signal residuals plus γ[(c1−c0)²+(c2−c1)²], with γ positive. The exact numerical weighting of residuals should reflect the measurement uncertainties in a real analysis.

Every feasible member of our family has zero signal residual. The uniform profile also has zero smoothing penalty, while the alternative has a positive penalty. The regularized fit therefore selects the uniform profile in this example.

That selection can be useful and transparent. It expresses a preference for a smooth profile among explanations supported by the data. It does not make the two signals distinguish the uniform profile from the alternative. To assess a conclusion that depends on the selected shape, vary the regularization assumption and examine the physical evidence for the preferred smoothness.

Contents · Previous chapter · Next chapter

Suppose independent physical reasoning requires the concentration to be nonincreasing with depth: c0 ≥ c1 ≥ c2. In our family, c0−c1 = 7ε/8 and c1−c2 = −7ε/4. The first inequality requires ε ≥ 0, while the second requires ε ≤ 0. Together they force ε = 0.

The uniform profile becomes the unique solution after adding that shape constraint. The data have not changed. What changed is the allowed set of profiles. A monotonicity assumption has supplied decisive information.

This example is not an argument against shape constraints. Such constraints can be appropriate when independently justified. It is a reason to name them. If a layered process could plausibly produce a buried enrichment or depletion, automatically imposing monotonicity might exclude the very feature the investigation seeks. The validity of the assumption belongs in the scientific evidence, not only in a fitting-software setting.

Contents · Previous chapter · Next chapter

Our two-profile ambiguity survives perfect angular positioning, noiseless intensities and exact calibration. It is an information limit of the selected forward problem. Better instruments can improve the supplied signals while leaving that particular limit intact.

Instrumental errors and model errors still matter. Uncertain angles perturb qθ; uncertain response factors change Sθ; roughness changes the path distribution. These effects can broaden or shift the feasible profile set. They should be added to the analysis rather than offered as explanations for an ambiguity already present without them.

A useful diagnostic sequence is to check exact identifiability first, then numerical conditioning, then measurement uncertainty and model adequacy. Otherwise a narrow numerical confidence interval can be mistaken for strong physical knowledge when it was generated inside an unjustifiably restrictive profile model.

Contents · Previous chapter · Next chapter

The simple exponential requires a transport parameter appropriate to the application. An inelastic mean free path and an effective attenuation length are related concepts with different uses; they should not be treated as interchangeable universal depth rulers.

Jablonski and Powell’s 2020 review of effective attenuation lengths explains that effective attenuation lengths account for elastic-scattering effects in specified quantitative XPS applications and depend on the defining equation and configuration. That supports checking the chosen transport description rather than assigning one unqualified length to every experiment.

Here λ is explicitly stipulated and constant so the inverse algebra can be isolated. For a real profile, establish whether that description is adequate for the materials, photoelectron energy and measurement geometry. If λ must also be inferred, the unknown space grows. The three-slab uniqueness result with known λ then no longer answers the expanded problem.

Contents · Previous chapter · Next chapter

A flat, laterally uniform model represents all points in the analysed area by the same depth coordinate and layer structure. Surface roughness or lateral patches can instead produce a distribution of local emission paths and compositions. A detector may average those contributions into one spectrum.

Fitting that average with a flat three-layer model can produce an apparently graded depth profile even when lateral structure is responsible for part of the signal variation. The numerical fit alone cannot establish that its one-dimensional description is the correct physical explanation.

Evidence about surface morphology, lateral uniformity and sample preparation therefore belongs beside the angular data. Use complementary observations to decide whether a one-dimensional reconstruction is an adequate question in the first place. Adding more adjustable depth slabs cannot automatically repair an incorrect representation of the sample geometry.

Contents · Previous chapter · Next chapter

Angle-resolved XPS changes the depth weighting while retaining the sample surface in the ideal measurement description. Sputter profiling removes material between observations. It can provide different depth information, but it also introduces its own preparation, depth-scale and alteration questions.

The existing SIMS and sputtering depth-profile article is useful for understanding that different evidence route. It should not be treated as an automatic ground-truth replacement for every angle-resolved reconstruction.

A complementary measurement is valuable when its sensitivity and limitations address the unresolved feature. State what new relationship it adds: a known boundary, an independent thickness, a chemical distinction or a different depth weighting. Naming a second technique without explaining the added constraint leaves the original inverse problem unresolved.

Contents · Previous chapter · Next chapter

Within our feasible family, a feature defined by a linear combination w0c0+w1c1+w2c2 is unchanged only if w0/8−3w1/4+w2 = 0. This is the condition that the feature ignore the unmeasured direction.

The two measured signals satisfy that condition by construction. The unweighted total, with w0 = w1 = w2 = 1, does not: its coefficient is 3/8. Nor does the first-slab concentration alone, whose coefficient is 1/8. These features cannot be reported as uniquely determined by the two readings.

This gives a practical route between claiming a full profile and giving up entirely. Ask which particular quantity the decision needs, then test whether that quantity is invariant across the feasible explanations. Data can determine some combinations accurately while leaving individual layers or inventories ambiguous. Reporting those combinations preserves useful information without overstating it.

Contents · Previous chapter · Next chapter

Murdoch and McCulloch’s 2023 Surface and Interface Analysis insight note describes angle-resolved XPS reconstruction as an inverse problem and discusses a Tikhonov regularization approach. Its introduction gives an exponentially weighted depth integral and states assumptions concerning sample geometry and electron transport.

That primary paper supports the measurement framework and the need to distinguish reconstruction assumptions from observations. The exact three-slab profiles, null direction, bounds and transfer exercises here are independently constructed examples. They are not profiles reported in that paper.

Both the method paper and the NIST review linked above were checked on 25 October 2026. The aim is to make a specialist inference readable without making it seem assumption-free. Use the source methods and appropriate experimental expertise when analysing actual spectra; use the worked algebra here to question what a particular set of signals can establish.

Contents · Previous chapter · Next chapter

Change the geometry to d = λ ln 3 while keeping the same three-slab definitions and known zero baseline. At normal emission r1 = 1/3, and at 60° r2 = 1/9. Begin with c = (0.3,0.3,0.3).

Calculate its two transformed signals. Then construct a null direction using the polynomial (r−1/3)(r−1/9). Add 0.3 times the coefficient vector of that polynomial to the original profile and verify that the resulting fractions remain physical and reproduce both readings.

Before calculating a third angle, predict whether the two profiles have equal total analyte inventory. Finally choose an angle with cos θ = 1/3, which makes r3 = 1/27 in this changed geometry. Calculate the difference of its transformed signals. This exercise changes the attenuation weights rather than merely halving every number in the original example.

Contents · Previous chapter · Next chapter

The baseline signals are S1 = 0.3(1+1/3+1/9) = 13/30, approximately 0.43333, and S2 = 0.3(1+1/9+1/81) = 91/270, approximately 0.33704.

The null polynomial is r²−4r/9+1/27, so its coefficient vector is (1/27,−4/9,1). Adding 0.3 times that vector gives (14/45,1/6,3/5), approximately (0.31111,0.16667,0.6). Substitution into both original signal equations leaves them unchanged. The fractions are all between zero and one.

The total concentration sum increases from 0.9 to 97/90, approximately 1.07778. At r3 = 1/27, the signal difference is 0.3(1/27−1/3)(1/27−1/9) = 8/1215, approximately 0.006584. A third weighting distinguishes the profiles, while the two original readings do not. The changed values preserve the same reasoning structure without relying on the original convenient half-power weights.

Contents · Previous chapter · Next chapter

Record the analyte signal, background and peak-area treatment, angle convention, response corrections, attenuation model and independent thickness information. State the permitted profile space, including the deeper baseline and fraction bounds. Those inputs define the inverse problem before any optimisation begins.

Then separate three results: the signal combinations actually constrained, the feasible range of scientifically relevant features, and the particular profile selected by any smoothing or shape assumptions. Test at least one alternative profile against the forward model when nonuniqueness is plausible.

Finally ask what independent observation would distinguish the alternatives that matter to the decision. A third angle, a known layer boundary or a complementary technique should be justified by its added sensitivity. A polished profile plot is easier to trust when its unresolved directions are also explained.

Contents · Previous chapter · Next chapter

Nonuniqueness does not mean that the readings contain no useful information. Subtracting the original two equations gives S0−S60 = c1/4+3c2/16 = 0.175. Equivalently, 4c1+3c2 = 2.8. This is a definite relationship between the middle and deeper concentrations, conditional on the supplied forward model.

Each candidate profile must also satisfy c0+c1/4+c2/16 = 0.525. These two relationships are the information carried by the two exact signals. They explain why increasing the deepest concentration requires decreasing the middle concentration and adjusting the surface concentration in the particular proportions already derived.

For a decision that depends only on one of these weighted combinations, the data may be sufficient. A decision that depends on the deepest concentration alone requires more information. This distinction is useful in communicating results: state the quantity actually constrained before displaying a preferred reconstruction. It avoids both overclaiming a full profile and discarding valid evidence because one requested feature remains unresolved.

Contents · Previous chapter · Next chapter

Our example supplies d and λ independently. If both were unknown, angular attenuation would depend on their ratio d/λ. Multiplying every depth, each slab thickness and λ by the same positive scale factor leaves rθ unchanged at every angle. For an analysis using angular shape with an unknown absolute response scale, these profiles cannot be distinguished by that shape alone. If Kθ is absolutely known, raw intensity magnitude may add a separate constraint; it must not be discarded by normalisation.

For example, two layer models with identical concentrations and identical d/λ produce the same transformed angular response even when one set of physical layer thicknesses is twice the other. This is a different ambiguity from the concentration null direction. Adding more angles can improve the angular constraints while leaving a missing external length calibration unresolved.

In a real analysis, an independently justified transport parameter or thickness measurement supplies information that angular shape does not automatically contain. Record which quantity anchors the length scale. Calling the reconstruction a depth profile should not conceal whether its horizontal axis is independently calibrated or inherited from an assumed attenuation length.

Contents · Previous chapter · Next chapter

At fixed d and λ, ln rθ = −(d/λ) sec θ. Its derivative with respect to θ in radians is −(d/λ) sec θ tan θ. The derivative vanishes at normal emission in this symmetric flat-surface model, but becomes larger in magnitude at more grazing angles.

For d/λ = ln 2 at 60°, the derivative is approximately −2.401 per radian. An angle error of one degree, about 0.01745 radians, gives a first-order change in ln r of approximately −0.04191. The weighting r therefore changes by roughly four percent in magnitude at this point. This is an illustrative sensitivity calculation, not a claimed instrument specification.

The resulting concentration uncertainty depends on the whole response matrix and calibration, so it should not be equated directly with four percent uncertainty in every layer. The calculation instead identifies an input that needs propagation. It also explains why moving to a grazing angle for stronger surface weighting can increase sensitivity to geometric errors at the same time.

Contents · Previous chapter · Next chapter

The first concentration c0 in our model is an average over a specified slab of nonzero thickness d. It is not an unrestricted point value c(0). In an integral measurement, changing a function at one isolated mathematical point changes no signal because that point has zero depth measure.

A finite altered surface layer does contribute. If its thickness is δ and its concentration differs by Δc, the magnitude of its intensity change in the restricted model is at most Kθ times the magnitude of Δc times δ, because the exponential weight is at most one. As δ becomes small, its integrated signal can become small even while its local concentration contrast remains large.

This observation does not imply that every thin surface feature is undetectable. It shows why detection, spatial resolution and a claimed local concentration must be tied to a finite representation and uncertainty. When reporting a surface concentration, state the layer thickness or averaging definition supported by the analysis. A numerical value at a plotted boundary should not silently acquire finer depth resolution than the measurements and model justify.

Contents · Previous chapter · Next chapter

Does a more grazing angle measure only the top slab?

No. It increases the relative weighting of shallow contributions in this model, but deeper terms remain nonzero. Calling it a direct top-layer concentration discards the weighted contributions that the forward equation still contains.

Do two signals always fail to determine any depth model?

No. A sufficiently restricted model with fewer independently unknown parameters can be identifiable. Our counterexample concerns three unknown slab concentrations with the other stated inputs known. The number and independence of the supplied constraints matter.

Does regularization make the reconstructed profile false?

It selects a profile using additional preferences or assumptions. That can be scientifically useful when those choices are justified and their influence is reported. The error is presenting the selected shape as uniquely demanded by data that admit alternatives.

Can three angles determine a continuous depth profile exactly?

The three-angle result here applies to three specified slab concentrations. A continuous profile has more freedom. Finite angular data generally require a finite representation, additional assumptions and uncertainty assessment rather than an unrestricted exact reconstruction.

Contents · Previous chapter · Next chapter

For the physical foundation, return to the X-ray Photoelectron Spectroscopy subject guide. For the distinction between surface signals and chemical-state conclusions, use XPS, surfaces and chemical-state evidence. The present article develops the separate question of whether angular signals uniquely specify a depth distribution.

For a different way of obtaining depth evidence, read SIMS, sputtering and depth-profile evidence, keeping its physical intervention and limitations explicit. Choose the route that addresses the unresolved feature rather than treating all surface-analysis measurements as interchangeable.

Connect this habit of testing alternative explanations through The Gold Standard of Education and the How X Works hub. A strong depth-profile conclusion states the weighting model, shows what the signals constrain and identifies the independent evidence behind any selected shape.

A useful check is to reconstruct each predicted raw signal after applying every stated response factor.

Contents · Previous chapter · Return to X-ray Photoelectron Spectroscopy