The Gold Standard of Visual Arts | Why Matching Opaque Paint Reflectance Does Not Determine the Same Mixture

Imagine an artist keeping two apparently interchangeable dark swatches. Each has been measured at the same wavelength, with the same appropriate geometry, after the layer is thick enough to hide its backing for the present comparison. The recorded diffuse reflectance is 0.50 in both cases.

This is a much stronger starting point than saying that two phone photographs look similar. Nevertheless, it remains a particular observation. It does not tell us how far light travels inside either material, how frequently its direction changes, or what a thinner layer would reveal.

The first learning move is to preserve the strength and the boundary of the observation together. We have stipulated a matching reflectance. We have not stipulated identical internal optical behaviour. A useful next experiment asks which difference could remain hidden by this measurement and whether that difference matters for the artistic decision.

Contents · Next chapter

Write R∞ for the model’s diffuse reflectance of an optically infinite layer at the selected wavelength. It is dimensionless. A value of 0.50 means a fraction, not fifty units of thickness or fifty units of absorption.

The infinity symbol describes a limiting optical condition. An actual layer need not have infinite physical thickness. The practical question is whether additional thickness or a different backing materially changes the measured result for the purpose and precision of the experiment. We are supplying the limiting condition in our mathematical examples.

A photograph, a display colour and a spectral reflectance are different records. A photograph also depends on illumination, receiving system and processing. Before applying an equation to a number, identify what the number represents. Calling any grey-looking image value R∞ would skip the very measurement condition that makes the following argument meaningful.

Contents · Previous chapter · Next chapter

Let K be the effective absorption coefficient and S the effective scattering coefficient in this particular two-flux model. We take K ≥ 0 and S > 0. Both use the same inverse-length unit, so K/S has no unit.

These are model coefficients. Do not automatically identify S with every scattering quantity used in a more detailed optical theory. The relationship between Kubelka–Munk coefficients and radiative-transfer quantities requires attention to the assumptions and illumination geometry. Thennadil's research paper examines that relationship.

For our fictional account, a larger K represents a larger absorption coefficient within the supplied convention, while S describes the competing scattering contribution. Neither coefficient alone is the observed reflectance. Keeping both symbols visible prevents an apparently dark sample from being assigned an absorption value that the observation never uniquely established.

Contents · Previous chapter · Next chapter

For the idealised opaque-layer model, define q = K/S. Its reflectance relationship is q = (1 − R∞)²/(2R∞), for 0 < R∞ ≤ 1. The Harrick explanation hosted by Caltech's materials research centre presents this transform and its semi-infinite-sample setting.

We will use the equation as a stated model, rather than treat it as a universal description of every studio surface. The original examples below investigate what this relationship can and cannot identify.

At R∞ = 0.50, the numerator is 0.25 and the denominator is 1. Therefore q = 0.25. We have learned that K is one quarter of S under the supplied model. We have not learned that K equals 0.25 in an inverse-length unit. A ratio and a coefficient are different quantities, even when the arithmetic looks simple.

Contents · Previous chapter · Next chapter

Choose one common inverse-length unit and consider three fictional samples. Their coefficients differ, but every ratio K/S is 0.25.

Check each division independently. The last sample has both coefficients ten times those of P. The middle sample has both one quarter of P’s coefficients. The opaque reflectance cannot distinguish them in this account.

The names are convenient labels, not recognised paint categories. Their purpose is to make the ambiguity concrete. If a worksheet reports only R∞ = 0.50 and then claims that the first row is the unique material explanation, the other rows are direct counterexamples. No measurement noise is needed to produce this failure of uniqueness.

Contents · Previous chapter · Next chapter

For any positive multiplier c, replacing K by cK and S by cS leaves their ratio unchanged. Substituting into the transform therefore leaves R∞ unchanged. There is a whole positive family of matching coefficient pairs.

More precise repetitions of the same ideal opaque reflectance may narrow measurement uncertainty about the ratio. They cannot remove this exact symmetry. It is possible to know q very precisely while remaining unable to identify K and S individually.

This is a useful distinction for an artist working with a technical collaborator. Ask whether the uncertainty comes from a noisy reading, an incomplete model or several exact material explanations of the same record. Each problem calls for a different response. Buying more precision does not automatically supply a new kind of information. The proposed measurement must distinguish the explanations that actually remain possible.

Contents · Previous chapter · Next chapter

To predict a reflectance after calculating q, rearrange the transform. Multiplying by 2R gives R² − 2(1 + q)R + 1 = 0. The physically relevant root within our stated interval is R = 1 + q − √(q² + 2q).

For q = 0.25, this becomes 1.25 − 0.75 = 0.50. The other algebraic root is 2, outside the permitted reflectance interval. Selecting a root requires the physical domain, not just an equation solver.

For numerical work at large q, the equivalent expression R = 1/[1 + q + √(q² + 2q)] can avoid subtracting two nearly equal large quantities. That is an arithmetic consideration inside the model. It does not improve the model’s description of a real sample. Keep numerical stability separate from physical adequacy when reviewing a prediction.

Contents · Previous chapter · Next chapter

Now suppose two compatible fictional materials are mixed with a supplied fraction f of A and 1 − f of B. For this exercise only, assume their effective coefficients add linearly: Kmix = fKA + (1 − f)KB and Smix = fSA + (1 − f)SB.

The fraction has the meaning stipulated by this model. We do not silently substitute the mass of wet paint, the number of brushfuls or a pigment-volume fraction from an unrelated formulation. Such quantities can differ.

This assumption is what allows us to make a mixture prediction. It is additional information beyond each unmixed opaque reflectance. Real formulations can change dispersion, packing, binder relationships and other conditions, so the additive account needs evidence before practical use. Our task is to examine its consequences accurately, then recognise where an actual material experiment must take over.

Contents · Previous chapter · Next chapter

Let A have KA = 0.50 and SA = 2.00. Its ratio is 0.25 and its opaque reflectance is 0.50. Let B have KB = 2.00 and SB = 2.00. Its ratio is 1 and its opaque reflectance is 2 − √3, approximately 0.268.

For an equal-fraction mixture, Kmix = (0.50 + 2.00)/2 = 1.25 and Smix = (2.00 + 2.00)/2 = 2.00. Therefore qmix = 0.625.

Using the inverse relationship gives Rmix = 1.625 − √1.640625, approximately 0.344. This is an opaque-layer prediction at the one selected wavelength, conditional on the supplied mixture rule. It is not a complete perceived colour and does not specify the appearance of a thin glaze on paper. Every qualification identifies a quantity the calculation has deliberately kept outside its boundary.

Contents · Previous chapter · Next chapter

Create B-prime with KB′ = 20 and SB′ = 20. Like B, it has ratio 1 and opaque reflectance approximately 0.268. The unmixed reflectance records cannot distinguish B from B-prime.

Mix A and B-prime in equal fractions under the same stipulated rule. Now Kmix = 10.25 and Smix = 11. The ratio is 41/44, approximately 0.932. Its predicted opaque reflectance is approximately 0.279.

The original mixture predicted about 0.344; the replacement predicts about 0.279. Both component pairs had identical individual opaque reflectances. Their mixtures differ because the hidden relative scattering scale changed. This is the central counterexample. A matching input swatch is insufficient evidence for an interchangeable mixing role, even before considering real-world departures from our simplified account. Ask which coefficients the substitution preserves, rather than assuming that one matching measurement preserves every consequence.

Contents · Previous chapter · Next chapter

Put the two hypotheses beside each other. Both use A with coefficients 0.50 and 2.00. Hypothesis one uses B with coefficients 2 and 2; hypothesis two uses B-prime with coefficients 20 and 20. Both explain the two supplied unmixed opaque reflectances exactly.

They disagree only when a question depends on the relative coefficient scale between components. The equal-fraction mixture supplies that question. The disagreement is therefore evidence about missing information, not a contradiction within the calculations.

This pattern transfers beyond painting: two models can fit the original observations and make different predictions for a changed condition. A useful experiment chooses a changed condition that separates the models. For this particular studio inquiry, the practical outcome is a reason to test the actual mixture rather than promise that a matching swatch guarantees an identical result.

Contents · Previous chapter · Next chapter

The arithmetic mean of the two original component reflectances is approximately (0.500 + 0.268)/2 = 0.384. That is neither of our two mixture predictions. It answers a different mathematical question.

In the supplied material model, we first combine K and S, then divide to obtain qmix, then convert qmix to reflectance. Averaging R values skips those relationships. The transform is nonlinear, so these operations do not generally commute.

An average reflectance can be meaningful for another explicitly defined arrangement, such as certain spatial averages of separate areas under suitable measurement conditions. That would be an account of an arrangement, not the coefficient-mixture rule used here. Before labelling a blended digital value as a physical paint mixture, identify which operation produced it. A number can be correct for one operation and misleading when carried into another without its conditions.

Contents · Previous chapter · Next chapter

Substitute KA = qASA and KB = qBSB into the mixture equation. Then qmix = [fSAqA + (1 − f)SBqB]/[fSA + (1 − f)SB]. The ratio is a weighted mean of the component ratios, with weights involving scattering as well as mixture fraction.

When SA = SB, those scattering factors cancel. Only in that special case does qmix equal fqA + (1 − f)qB. Even then, reflectance is obtained afterwards through the nonlinear inverse relationship.

When B’s scattering coefficient is ten times A’s, equal component fractions do not give equal weights in this ratio. B contributes ten elevenths of the effective ratio weighting. This explains the counterexample in words: the opaque swatch recorded B’s ratio, while its stronger coefficient scale makes it dominate the model’s mixed ratio.

Contents · Previous chapter · Next chapter

Because the mixture weights are nonnegative, qmix lies between qA and qB whenever the denominator is positive. In our example it lies between 0.25 and 1. As R decreases with q, the mixed opaque reflectance lies between approximately 0.268 and 0.500.

This bound is weaker than a unique prediction, but it is still information. Unknown relative scattering does not mean that absolutely anything is possible under the supplied model. It means the precise position within this conditional range remains unresolved.

Check the assumptions before using the bound as evidence. It depends on the additive-coefficient model and positive coefficients, not just on two observed swatches. A real mixture outside the predicted interval could indicate unsuitable assumptions, measurement differences or an incomplete material account. It would not justify quietly changing a recorded value until the bound appears to hold.

Contents · Previous chapter · Next chapter

Keep the original A and B with equal scattering coefficients. At f = 0.75, qmix = 0.75 × 0.25 + 0.25 × 1 = 0.4375. Converting that ratio gives an opaque reflectance of approximately 0.405.

Now use B-prime instead. The numerator for qmix is 0.75 × 0.50 + 0.25 × 20 = 5.375, and the denominator is 0.75 × 2 + 0.25 × 20 = 6.5. Thus qmix = 43/52, approximately 0.827, giving a reflectance of approximately 0.298.

The larger fraction of A moves both mixtures towards A’s reflectance, but it does not make the two predictions equal. A coefficient ledger lets the learner explain this movement without guessing from a colour name. The result also shows why recording the actual supplied fraction matters: a vague instruction to add more of one colour cannot be checked against a quantitative model.

Contents · Previous chapter · Next chapter

Suppose the additive model has independent support and an equal-fraction opaque mixture has also been measured. Write t = SB/SA. Then qmix = (qA + tqB)/(1 + t). Rearranging gives t = (qmix − qA)/(qB − qmix), provided the denominator is nonzero.

With qA = 0.25, qB = 1 and qmix = 0.625, this gives t = 0.375/0.375 = 1. With qmix = 41/44, it gives t = 10. The extra mixture record distinguishes the two relative scattering explanations.

This is a conditional identification result. It relies on a known mixture fraction, a consistent optical measurement and a suitable mixture model. It does not follow merely because a third attractive swatch has been photographed. The additional record must connect to the unresolved parameter through a justified relationship.

Contents · Previous chapter · Next chapter

Even after t is identified, multiplying every K and every S in the whole material ledger by the same positive factor leaves every additive opaque-mixture reflectance unchanged. Component ratios remain unchanged, relative scattering remains unchanged, and all mixture ratios remain unchanged.

Therefore a large collection of opaque mixture records can still leave a common absolute coefficient scale unresolved. The records can improve useful mixture prediction within the model while failing to identify a length scale.

This is not a reason to dismiss the experiments. It is a reason to match the achievement to the question. An artist may need a dependable conditional mixture comparison rather than an absolute optical coefficient. A researcher asking about thickness-dependent behaviour needs another kind of information. State which task the collected measurements support before describing the material as fully characterised.

Contents · Previous chapter · Next chapter

If an independent, applicable measurement establishes SA = 2 in the chosen inverse-length unit, then qA = 0.25 gives KA = 0.50. If the mixture has also identified t = 10, SB = 20 and KB = 20 follow.

The word applicable carries real work. The independent coefficient must use the same model convention and relevant material conditions. A scattering quantity from another theory, a manufacturer description of opacity or a visual estimate of coverage cannot simply be inserted as S.

Caldwell's original study discusses obtaining Kubelka–Munk coefficients from transmittance measurements of specimens with a stated thickness relationship. It provides a research route into additional information, not a promise that one casual studio measurement supplies both coefficients. Follow a suitable validated method when actual optical characterisation is required.

Contents · Previous chapter · Next chapter

All numerical examples so far concern one wavelength. A visible spectrum contains many wavelengths, and K and S can depend on wavelength. A complete spectral ledger would preserve that dependence instead of assigning one ratio to an entire colour.

Two materials can share an opaque reflectance spectrum while retaining coefficient-scale ambiguities at the measured wavelengths. Conversely, two samples that match only three receiving-channel readings may already have different reflectance spectra. These are different layers of missing information.

The existing guide to matching colour readings and hidden reflectance spectra develops the receiving-channel problem. Here we have granted a reflectance at the wavelength under discussion and examined the material explanations beneath it. Knowing which layer is unresolved prevents one additional colour reading from being credited with identifying every hidden material property.

Contents · Previous chapter · Next chapter

Our mixture rule combines coefficients for a fictional homogeneous mixture. It does not describe two separate layers applied one above the other. Layer order, thickness, backing and surface boundaries can become important in a layered account.

A thin transparent-looking wash also cannot be treated as the opaque limit merely because its colour has been named. If light reaches the backing and returns through the layer, the backing remains part of the appearance. Supplying R∞ where a finite-layer observation belongs changes the problem.

For practice, label three sketches: separate adjacent areas, a homogeneous mixture, and one layer over another. Write which mathematical operation has been supplied for each. If only the mixture operation is defined, leave the other two predictions unresolved. This discipline makes the model useful without asking it to answer a different physical question.

Contents · Previous chapter · Next chapter

Suppose a hypothetical opaque reflectance is known only to lie from 0.48 to 0.52. The transform decreases over this interval. Its ratio therefore lies from F(0.52) to F(0.48), where F(R) = (1 − R)²/(2R).

The endpoints give approximately 0.2215 and 0.2817. The interval is not obtained by simply adding and subtracting 0.02 from the central ratio 0.25. The measurement and transformed quantity have different scales.

Without an absolute S reference, these ratio bounds still do not create finite individual bounds for K and S. For each permitted positive ratio, scaling both coefficients together preserves the observation. A responsible uncertainty account should retain both facts: the ratio has a conditional interval, and the individual coefficient scale remains unidentified. A narrow error bar around an arbitrarily selected pair would hide that second limitation.

Contents · Previous chapter · Next chapter

The formula t = (qmix − qA)/(qB − qmix) becomes sensitive when qmix approaches qB. The denominator becomes small. A modest change in the inferred mixture ratio can then produce a large change in the relative scattering estimate.

This is a different issue from exact non-identification. Away from an endpoint, an exact suitable mixture record may identify t. Near the endpoint, realistic uncertainty can make the estimate poorly constrained. The algebra should guide the experiment design rather than encourage reporting more digits.

For an actual investigation, a collaborator would need to assess measurement precision, suitable mixture fractions and model adequacy together. Our paper exercise supplies no universal choice of fraction or instrument. Its contribution is to make the question precise: which proposed record separates the remaining relative-scale possibilities by more than the relevant uncertainty?

Contents · Previous chapter · Next chapter

Even a well-supported pair of effective coefficients does not automatically identify a unique pigment, binder or formulation. Different underlying structures can have similar effective behaviour within a particular model and measurement range.

Likewise, a prediction about optical appearance does not establish lightfastness, handling, adhesion, toxicity or long-term compatibility. Those are separate material and studio questions. A material can match a colour target while being unsuitable for the intended surface or learning setting.

The practical lesson is modest and useful. Keep optical evidence with the optical claim, and obtain the relevant product information and appropriate expertise for other decisions. A learner does not need to improvise with unknown powders or solvents to understand the mathematics. Fictional ledgers and ordinary approved studio comparisons are enough to reveal the hidden assumption behind an interchangeable-swatch claim.

Contents · Previous chapter · Next chapter

Start with an artistic question that can actually be observed: will a replacement material produce the intended mixture relationship under the chosen conditions? Keep the support, preparation, mixture convention, drying condition, lighting and comparison method documented as appropriate to the activity.

Use the products and methods permitted in the learning setting. Record what was supplied, what was changed and what was observed. If the purpose is visual exploration rather than optical characterisation, say so. A careful qualitative comparison can still guide the next version without being relabelled a coefficient measurement.

The model suggests why testing the mixture can be more informative than repeatedly inspecting unmixed swatches. It does not require every art exercise to become a laboratory procedure. Choose the smallest comparison that answers the real creative question, and keep the result available for the next changed material or composition.

Contents · Previous chapter · Next chapter

A fictional learner writes: both paints reflect half the light, so their absorption coefficients are equal. Ask them to identify the missing variable before correcting the sentence.

Under our stated opaque-layer relationship, each reflectance gives K/S = 0.25. The learner can then construct K = 0.50, S = 2 and K = 5, S = 20. Both explain the same record, so equality of K was never established.

Next ask the learner to make the revised statement useful: the measurements establish the same ratio in this model; mixture prediction additionally needs relative scattering information and an applicable mixture rule. This repair preserves the valid observation while removing the unsupported conclusion. It is more informative than merely crossing out the answer or saying that real paint is complicated. The learner now knows exactly which evidence would change the claim.

Contents · Previous chapter · Next chapter

Use a fresh invented pair. Material C has KC = 1 and SC = 2, so qC = 0.50. Material D has KD = 6 and SD = 3, so qD = 2. Apply the same equal-fraction additive-coefficient rule.

Before calculating, answer three questions. Must the mixture ratio be the arithmetic mean of 0.50 and 2? Could D be replaced by a material with KD′ = 12 and SD′ = 6 without changing its unmixed opaque reflectance? Would that replacement necessarily preserve the mixed reflectance?

Work from the coefficient ledger and convert to reflectance only at the end. Then explain which conclusion comes from the opaque transform and which comes from the additional mixture assumption. This changed context checks whether the reasoning transfers beyond the convenient equal-scattering values in the first example. An unexplained memorised average is not enough to answer it.

Contents · Previous chapter · Next chapter

For C and D, the equal-fraction mixture has K = 3.5 and S = 2.5, so q = 1.4. Its reflectance is 2.4 − √4.76, approximately 0.218. The arithmetic mean of the component ratios is 1.25, a different value.

D-prime keeps qD′ = 2 and therefore keeps D’s opaque reflectance. But its mixture with C has K = 6.5 and S = 4, so q = 1.625. Its reflectance is 2.625 − √5.890625, approximately 0.198. The replacement changes the conditional mixture prediction.

For an extension, multiply all four original C and D coefficients by three. The mixed ratio remains 1.4. Explain why scaling one component can matter while scaling the whole ledger together does not. This comparison reveals the distinction between relative coefficient information and a common absolute scale.

Contents · Previous chapter · Next chapter

Suppose both components have q = 0.25, although their individual scattering coefficients differ. Substituting KA = 0.25SA and KB = 0.25SB makes Kmix exactly 0.25Smix for every permitted fraction. The opaque mixture therefore retains R = 0.50 in this model.

This special case matters. Our counterexample does not prove that every hidden coefficient difference must change every mixture. It proves that the original observations alone do not guarantee interchangeability in the changed mixture we constructed. A strong explanation distinguishes a possible consequential difference from an inevitable one.

Ask the learner to identify the algebraic cancellation before introducing another numerical example. Recognising when missing information does not affect a particular answer is as useful as detecting when it does.

Contents · Previous chapter · Next chapter

An equal-fraction experiment identified t because its fraction was supplied. If f is also unknown, the general ratio depends on the relative weight w = (1 − f)SB/(fSA), for 0 < f < 1. The mixture record can identify w under suitable conditions without identifying f and SB/SA separately.

For example, w = 1 can arise from equal fractions and equal scattering, or from f = 0.80 with SB/SA = 4. Both produce equal effective ratio weights. A vaguely recorded mixture cannot be treated as a precise calibration of relative scattering.

This supplies a practical reason to preserve the preparation record. A missing fraction is not merely an untidy notebook entry; it changes which material quantities the observation can identify.

Contents · Previous chapter · Next chapter

The transform divides by R. An exactly zero reflectance cannot simply be entered into it as a finite coefficient ratio. In the model, reflectance approaches zero as q grows without bound. A recorded zero may also involve a measurement limit, which needs its own account.

At the other endpoint, R = 1 gives q = 0. With S > 0, this implies K = 0 in the stated model, while S remains free. Even this unusually informative absorption result does not identify both coefficients.

These endpoint checks prevent a familiar error: treating a formula’s printed symbols as permission to use every possible display value. Identify the mathematical domain and the measurement meaning together before computing a transformed quantity.

Contents · Previous chapter · Next chapter

Imagine that the fictional additive account predicts 0.344 but a properly characterised actual mixture gives a materially different result. First reproduce the calculation from the original ledger. Then check whether the supplied coefficients, fraction and measurement conditions describe the same material experiment.

If those records agree, investigate model adequacy rather than editing the observation to preserve the prediction. A coefficient model can be mathematically consistent and still omit an important material behaviour. Conversely, a sound model can be given the wrong units or mixture fraction.

Keep these possibilities separate in the notebook. The next useful action depends on where the disagreement enters: arithmetic, records, measurement or assumptions. That makes an unexpected result a source of learning instead of an excuse to report only the successful swatch.

Contents · Previous chapter · Next chapter

The repair route starts with one ratio and two coefficient pairs. Ask the learner to explain why a matching opaque reflectance does not identify either coefficient separately. Keep the mixture calculation aside until that distinction is stable.

The stabilisation route adds the coefficient table, an equal-fraction mixture and a substituted opaque match. Change one condition at a time. The learner should be able to say why the predicted mixture changed, not merely produce two decimal answers.

The extension route uses unequal fractions, relative-scale recovery and measurement intervals. It also asks when the model is inappropriate. Independence appears when the learner chooses which record is needed for the actual question and acknowledges a conclusion the evidence cannot support. The Education parent guide connects that habit with learning that remains usable in a new situation.

Contents · Previous chapter · Next chapter

Does a matching opaque reflectance make two paints interchangeable?

It establishes the stated optical match. Our counterexample shows that it does not, by itself, establish the same mixture prediction under the supplied coefficient model. Other practical properties require their own evidence too.

Can I average the two reflectances?

You can calculate their arithmetic mean, but that operation is not the homogeneous coefficient-mixture rule used here. Identify the arrangement and model before interpreting the average as a material prediction.

Does the model tell me which commercial paint to buy?

No product comparison has been performed. The numerical materials are invented. Use appropriate product information, actual relevant comparisons and suitable advice for the intended work.

Would a complete reflectance spectrum solve everything?

It would provide more spectral information than a few colour readings. It would still need an applicable material model and additional evidence to identify every absorption and scattering coefficient separately.

Contents · Previous chapter · Next chapter

For artistic purpose and controlled changes, return to Visual Arts. For the separate receiving-channel ambiguity, use matching colour readings and hidden spectra.

For practical media, continue with Gouache Painting, Acrylic Painting or Watercolour Painting. Their handling and making questions should retain their own material context.

For research context, Haase and Meyer's publication record describes applying Kubelka–Munk pigment mixing to image synthesis. The Caltech-hosted Harrick note supplies the opaque transform; Thennadil and Caldwell supply the distinct research routes identified above. Sources checked on 24 October 2026. All numerical ledgers, substitutions, bounds and transfer tasks here are original hypothetical constructions. Explore How X Works for further connections.

Contents · Previous chapter · Return to Visual Arts