The Gold Standard of X-ray Science | Recovering Detector Dead Time from Known Input Ratios

A displayed count rate can be smaller than the rate entering a detector’s counting process. If both the input rate and the detector’s recovery parameter are unknown, one displayed value does not normally identify both. A plausible correction needs the missing information rather than a guessed parameter.

Let S denote the reference pre-dead-time event rate. Let τ denote the recovery time in the supplied response model. These symbols describe two different properties: S concerns the input condition, while τ concerns the modelled detector response. Neither is directly the displayed rate.

Now suppose a second supplied condition has a pre-dead-time rate known to be rS, where r is a positive fraction below one. The same response parameter applies to both conditions. That known relationship is additional information; it is not inferred merely because someone intended to reduce the source.

The question becomes concrete: given the two recorded means and r, which S and τ reproduce both? This is different from correcting readings when τ is already known. We will derive the parameters, check their physical domain and test a third condition. A successful pair fit is a conditional mathematical result, not proof that every assumption describes an actual detector.

Contents · Next chapter

Use the supplied stationary mean-rate relationship m = n/(1 + nτ). Here n is the eligible event rate before this particular dead-time loss, and m is the recorded mean rate. The nonparalyzable idealisation means missed events during a recovery interval do not extend that interval.

If rates are measured per second, τ must be in seconds. The product nτ is dimensionless. The equation does not also correct detection efficiency, changing energy response, geometry or every other effect in the measurement chain.

The specialist photon-counting detector primer, first published on 21 July 2026, gives context for classical throughput models and their limits. Our use of its response relationship is explicit; the input ratios and numerical records below are invented teaching assumptions.

Keep the word mean attached to m. The exact examples supply model means, so their equations can be solved exactly. Real finite observations vary, and detector output is not automatically a simple independent Poisson process. Estimation, uncertainty and model validation are additional jobs. A classroom inverse should clarify those boundaries rather than suggest that two noisy numbers can certify a response model without further evidence.

Contents · Previous chapter · Next chapter

For a single recorded mean a, the inverse relationship is S = a/(1 − aτ) when 1 − aτ is positive. Different allowable values of τ produce different values of S while reproducing the same a through the forward model.

Take a fictional recorded mean of forty events per second. With τ = 0, the required input is forty. With τ = 0.01 seconds, it is forty divided by 0.6, or 66⅔. With τ = 0.02 seconds, it is forty divided by 0.2, or two hundred.

Each pair gives forty when substituted into m = n/(1 + nτ). The observation alone does not choose among them. Averaging their S values would not repair the missing parameter; it would average answers based on different assumptions without a supplied reason for doing so.

A useful learning check asks for two compatible parameter pairs before adding more data. The learner then sees what the second condition must accomplish: restrict a family, rather than simply provide another number to average. The question concerns identifiability under a model. More decimal places in the first reading do not create the missing independent relationship between input and response.

Contents · Previous chapter · Next chapter

Let a be the reference recorded mean from input S. Let b be the recorded mean from input rS. Under the shared response, a = S/(1 + Sτ) and b = rS/(1 + rSτ). The known r belongs to the pre-dead-time inputs, not to the displayed readings.

For the first fictional record, set r = 0.5, a = 40 and b = 25 events per second. The supplied second input is half the first. The displayed second reading is twenty-five fortieths, or 0.625, of the first.

Do not treat 0.625 as a failed statement of the supplied input ratio. The response is nonlinear. The lower input condition experiences a different loss fraction under the same model.

Also avoid claiming that a physical attenuator necessarily supplies r = 0.5 for every relevant event. Changes in spectrum, geometry, background or detection efficiency can affect that relationship. Here r is an independently supplied property of the eligible pre-loss event stream. Establishing it in real work would require appropriate evidence and competent methods beyond this exercise.

Contents · Previous chapter · Next chapter

Take reciprocals of the two positive recorded means. From a = S/(1 + Sτ), obtain 1/a = 1/S + τ. From b = rS/(1 + rSτ), obtain 1/b = 1/(rS) + τ.

The transformed equations are simpler because the shared τ now appears as the same additive term. Their difference removes it: 1/b − 1/a = (1 − r)/(rS). The difference is positive for a finite positive S and a supplied r below one.

In the original pair, 1/25 − 1/40 = 0.04 − 0.025 = 0.015 seconds. Since (1 − r)/r equals one when r = 0.5, this reciprocal difference is exactly 1/S. Therefore S = 1/0.015 = 66⅔ events per second.

Explain the units before proceeding. A reciprocal rate has units of time per counted event; counts are treated as dimensionless in this model, leaving a time unit compatible with τ. The transformation changes the representation of the data. It does not make a noisy reciprocal an unbiased observation or remove the need for a suitable fitting method in an actual measurement problem.

Contents · Previous chapter · Next chapter

Once S is known, use τ = 1/a − 1/S. For the original pair, τ = 0.025 − 0.015 = 0.01 seconds. Both unknowns have now been determined under the supplied model and input ratio.

An equivalent direct expression is τ = (1/a − r/b)/(1 − r). Substituting a = 40, b = 25 and r = 0.5 gives (0.025 − 0.020)/0.5 = 0.01. This provides a second algebraic route for checking the result.

The input scale can also be written S = ab(1 − r)/[r(a − b)]. This form makes the danger of a small recorded-rate difference visible in the denominator. The reciprocal form makes the shared additive parameter visible. Each representation highlights a different feature of the same inverse.

Keep both unknowns in the conclusion: the fictional reference input is 66⅔ events per second and the fitted recovery parameter is 0.01 seconds. Reporting only one can hide the fact that they were estimated together from the same supplied pair. Neither number is a general property of X-ray detectors, a recommended setting or an independently verified property of real equipment.

Contents · Previous chapter · Next chapter

Pass S = 66⅔ and τ = 0.01 through the reference response. The denominator is 1 + ⅔ = 1⅔, giving forty. For the reduced condition, the input is half of 66⅔, or 33⅓. Its denominator is 1 + ⅓ = 1⅓, giving twenty-five.

Both supplied recorded means are reproduced. This is the essential forward check. A result that fits only the reference condition would leave the second observation unexplained, even if the recovered parameter looked plausible in isolation.

Try an alternative assumption of τ = 0. The reference reading would require S = 40. A half input would then give twenty, not the supplied twenty-five. The second record rules out that zero-dead-time pair within the exact exercise.

This check also clarifies the limits of the result. Two equations with two unknowns can produce an exact pair fit because the model was chosen to allow it. Reproduction of the two inputs confirms the algebra and pair compatibility. It does not establish that the response remains valid at other rates, that r was measured correctly or that an omitted background term is absent. Those questions need additional evidence.

Contents · Previous chapter · Next chapter

Let q = b/a, the ratio of the two recorded means. Dividing the forward equations gives q = r(1 + Sτ)/(1 + rSτ). For nonnegative τ, positive finite S and 0 < r < 1, q must lie between r and one, including r but excluding one.

At τ = 0, the response is linear and q = r. With positive τ, the larger input loses a greater proportion, so the displayed ratio moves above r. It remains below one for a finite positive input under the stated model.

The original q is 0.625, which lies above 0.5 and below one. The ratio therefore passes this necessary domain check. It is more informative than checking only that both recorded means are positive.

This is a model-compatibility statement, not a universal detector law. A different response, changing input conditions or finite measurement fluctuations can yield other observed ratios. When an exact exercise violates the interval, identify that violation directly. When an actual noisy dataset does so, investigate its uncertainty and assumptions rather than forcing a negative physical recovery time into the interpretation.

Contents · Previous chapter · Next chapter

Keep the known input ratio r = 0.5 and reference recorded mean a = 40, but change b to sixteen. The recorded ratio is 0.4, below the supplied input ratio. The pair cannot be reproduced by a finite positive input and nonnegative τ in this model.

The algebra exposes the problem. The direct expression gives τ = (0.025 − 0.5/16)/0.5 = −0.0125 seconds. The corresponding S is 26⅔, smaller than the reference recorded mean despite the model representing count loss. The mathematical solution has left the permitted physical domain.

Do not fix this by deleting the minus sign. That would create a different parameter pair that no longer solves the supplied equations. Nor does the negative result prove that a particular instrument is defective; the fictional inputs might be inconsistent because a premise was changed.

Write a useful conclusion: under the stated exact model and known half-input relationship, these readings have no solution with nonnegative recovery time. In an actual investigation, relevant possibilities include rate uncertainty, an incorrect input ratio, changing conditions or an inadequate response description. The exercise identifies the contradiction; it does not select its physical cause without evidence.

Contents · Previous chapter · Next chapter

If r = 0.5, a = 40 and b = 20, the recorded ratio equals the input ratio. The fitted τ is zero and S is forty. This is the permitted linear boundary of the model, not a reason to claim every real recovery interval has been demonstrated absent.

Now set a and b both to forty while retaining r = 0.5. For a finite positive S, the known inputs differ, and the supplied response is strictly increasing. Equal positive recorded means cannot be produced exactly by that finite pair under the model.

The formula for S reflects the boundary through a zero a minus b denominator. Reporting an enormous finite value after rounding away that zero would disguise the identifiability problem. As the two readings approach equality, compatible inferred S values can grow without bound while τ approaches the reciprocal of a.

These cases are useful checks for a calculation sheet. The linear pair should yield τ = 0. The equal positive pair should be flagged as lacking a finite solution under the premises. Treating both as ordinary numerical outputs would hide the distinction between a valid boundary value and an unavailable finite inverse.

Contents · Previous chapter · Next chapter

The original pair gave S = 66⅔ and τ = 0.01. Suppose a third supplied condition has one quarter of the reference pre-loss input, so r_3 = 0.25. The model predicts input 16⅔ and recorded mean 100/7, approximately 14.286 events per second.

This third value was not used to solve the first pair. It therefore checks a consequence of the fitted model rather than merely repeating the fitting arithmetic. If an exact third record instead gives fifteen, the three exact conditions are not all reproduced by the same pair S and τ.

That discrepancy does not automatically identify which premise failed. It could concern the third ratio, the stable source assumption, the common response or another omitted effect. With noisy measurements, a residual should be interpreted through an appropriate uncertainty and fitting framework rather than treated as an exact contradiction.

The practical learning habit is to reserve a changed condition when possible. A formula that reproduces the data used to construct it has achieved less than a model that also predicts independent evidence within a defensible scope. Even several matching checks do not establish unrestricted validity, but they make the support more informative than one successful pair fit.

Contents · Previous chapter · Next chapter

For a relative input r, the model gives 1/m = τ + (1/S)(1/r). If x = 1/r and y = 1/m, the supplied exact records lie on a straight line with intercept τ and slope 1/S.

The original conditions give x values one and two, with y values 0.025 and 0.040 seconds. Their slope is 0.015 and intercept 0.010. The quarter-input prediction adds x = 4 and y = 0.070, consistent with the same line.

This representation helps a learner distinguish input ratio from displayed ratio and recognise what a third record tests. It does not mean that an ordinary unweighted straight-line fit to reciprocals is automatically the right estimator for noisy count data. Transformations change uncertainty and can make errors unequal across points.

For a real analysis, retain the original counts, time definitions and relative-input evidence, and use a method suited to their statistical model. For the exact classroom record, the line is an inspectable representation of the algebra. A straight-looking plot is not proof that the physical setup obeys every premise or that a calibration may be extrapolated beyond its supported conditions.

Contents · Previous chapter · Next chapter

The expression S = ab(1 − r)/[r(a − b)] contains a minus b in its denominator. When two recorded means become close while r remains meaningfully below one, small changes in their difference can cause large changes in the inferred input scale.

Keep a = 40 and r = 0.5. If b = 25, S is 66⅔. If b = 36, S is 360 and τ is approximately 0.02222 seconds. If b = 39.6, S is 3,960 and τ is approximately 0.024747 seconds.

All three fictional pairs have the same reference recorded mean, but their different second readings imply very different joint parameter estimates. As b approaches forty from below, the inferred input scale grows while the recovery parameter approaches 0.025 seconds.

The lesson is not that close readings are always bad. They may be appropriate evidence for another question. Here, they make this particular inverse sensitive because the supplied model approaches a saturating response. A report should explain that sensitivity rather than present a very large inferred S with many digits as if the digits alone established precision.

Contents · Previous chapter · Next chapter

The known r is essential to the solution. If it is merely assumed, uncertainty in that assumption travels into both recovered parameters. Two recorded means do not independently verify the relative input that the calculation uses.

Hold the recorded pair at forty and twenty-five. With r = 0.5, the solution was S = 66⅔ and τ = 0.01. With r = 0.4, the solution becomes S = 100 and τ = 0.015. With r = 0.625, the linear solution S = 40 and τ = 0 also reproduces the pair.

These are different supplied input relationships, not contradictory solutions under one unchanged r. They demonstrate what is lost when the input ratio is unknown. The recorded ratio 0.625 cannot simply be substituted as r without imposing the linear interpretation one hoped to investigate.

Preserve how the ratio was established and what it refers to. A nominal change in a control, a geometric distance or a material thickness is not automatically the same as a known ratio of eligible pre-dead-time events. The relationship needs appropriate physical evidence before the algebra can support an instrument conclusion.

Contents · Previous chapter · Next chapter

A statement that a material transmits half of some incident radiation does not necessarily establish r = 0.5 for the event stream used in this model. Energy-dependent transmission and detector response can change the eligible event distribution. Background may not scale in the same way as the desired source.

That is why the teaching problem supplies the relative input directly instead of instructing a reader to insert an object into a beam. The article concerns inference from a declared record. It does not specify how to create those conditions or authorise an equipment experiment.

In X-ray work, keep the chain visible: incident field, interaction, eligible detector events, counting response and reported quantity. A known relationship at one stage cannot be silently transferred to another stage without support.

The existing guide on why one transmission reading cannot identify a material develops related ambiguity about the physical interpretation of a transmission value. Here the missing link can be the relative pre-loss rate itself. A mathematically exact parameter estimate remains conditional on that link being correctly established.

Contents · Previous chapter · Next chapter

The main model assumes n(r) = rS with no additional background term. If the pre-dead-time field instead has n(r) = rS + B for an unknown unscaled background B, the response becomes m(r) = (rS + B)/[1 + τ(rS + B)].

There are now three unknown parameters: S, B and τ. The two-equation formulas derived earlier do not apply unchanged. Applying them anyway can produce numbers that fit a simplified pair while misrepresenting the components of the actual field.

For an original counterexample, let S = 60, B = 10 and τ = 0.01. The reference input is seventy and the half-source input is forty. Their recorded means are approximately 41.18 and 28.57, whose ratio is not generated by simply halving the total pre-loss input.

The existing Photonics correction-order guide explains comparison of gross and background acquisitions when the response parameter is supplied. That is a separate problem. This article estimates an unknown response parameter from a known relative-input relationship. Identifying which problem the record supports prevents a correct formula from being used with the wrong number of unknowns.

Contents · Previous chapter · Next chapter

Before forming a recorded rate, establish what the supplied count and duration mean. Clock time, reported live time and an already corrected rate can belong to different processing conventions. A label of events per second does not by itself identify the scale used by the model.

For a fictional record, 2,400 events over sixty seconds gives forty per second under the exercise’s declared elapsed-time basis. A second record of 750 over thirty seconds gives twenty-five per second on the same basis. Subtracting raw counts or assuming equal durations would not reproduce those inputs.

If the system already reports a dead-time-corrected estimate, feeding it into a formula derived for raw recorded means would change the problem. Retain the processing label and relevant acquisition definitions with the values.

The X-ray Spectroscopy guide introduces rate and duration distinctions for background comparisons. Here those definitions are prerequisites to parameter recovery. Correct algebra should begin only after the observations have been placed on the declared scale. Otherwise, the apparent fit may explain a mixture of incompatible quantities rather than a detector response.

Contents · Previous chapter · Next chapter

The exercise solves supplied mean rates exactly. Actual count records have uncertainty, and the nonlinear inverse can amplify it. Uncertainty in the reference and reduced readings, their dependence, the input ratio and the response assumptions can all matter.

Near the linear boundary, a small change in the reduced reading can move the exact fitted τ through zero. With a = 40 and r = 0.5, b = 20 gives zero. Values just above twenty give a small positive fitted parameter; values just below give a negative exact fit outside the allowed domain.

That sensitivity is not repaired by silently truncating every negative estimate to zero and then claiming the original estimator was unchanged. A constrained statistical analysis is a distinct method that needs an appropriate interpretation. The particle-counter mathematics guide supplies a route into counting models and their limits.

For a classroom report, separate the exact parameter solution, a stated sensitivity exercise and an actual uncertainty estimate. Use confidence language only when a valid inferential method supports it. This preserves the value of the algebra without asking it to answer statistical questions for which the exercise supplied no probability model.

Contents · Previous chapter · Next chapter

A fictional student, Arun, sees the known half-input condition and expects the display to fall from forty to twenty. When it falls to twenty-five, he concludes that the supplied ratio must be wrong. His first unstable step is assuming that the detector response is linear.

Start with the forward model using a supplied S and τ. Have him calculate the reference and half-input displays before attempting to infer parameters. The two different loss fractions become visible, and the recorded ratio can be compared with the pre-loss ratio without treating them as the same quantity.

Next, cover the parameter values and supply the two mean readings. Arun writes the reciprocal equations, identifies the shared additive term and removes it by subtraction. He then checks the recovered values forward rather than relying on a remembered formula alone.

Finish with an incompatible ratio and ask for an explanation rather than a positive number at any cost. A learner who can state that the exact premises have no permitted solution has gained a useful scientific capability. The goal is increasing control over the model and its limits, not automatic production of a calibration-shaped answer for every pair of inputs.

Contents · Previous chapter · Next chapter

Use a new fictional case with r = 0.25. The reference recorded mean is exactly 400/7 events per second, approximately 57.143. The reduced-condition recorded mean is exactly 200/11, approximately 18.182. Use the fractions for the calculation rather than the rounded displays.

Recover S and τ under the same nonparalyzable model and no-background premise. Check the recorded-ratio domain. Then reconstruct both supplied mean readings through the forward response and predict the recorded mean at r = 0.5.

Now replace the known quarter-input relationship with an unknown r. Explain why the two observed means alone no longer identify the same pair of parameters. Do not use the displayed ratio as a substitute without explaining the new assumption that would impose.

Finally, suppose a third exact half-input record is supplied and differs from your prediction. State what is inconsistent and what remains unknown about the cause. This exercise changes the relative input and the arithmetic while preserving the reasoning structure. Write a conclusion naming the recovered scale, model parameter, independent check and the condition that gives the result its physical meaning.

Contents · Previous chapter · Next chapter

The reciprocal rates are 7/400 = 0.0175 and 11/200 = 0.055 seconds. Their difference is 0.0375. With r = 0.25, S = 0.75/(0.25 × 0.0375) = eighty events per second.

The response parameter is 0.0175 − 1/80 = 0.005 seconds. The recorded ratio is 7/22, approximately 0.3182, which lies between the supplied 0.25 and one. The domain check agrees with the nonnegative recovered parameter.

For the independent half-input check, forty divided by 1.2 gives 100/3, approximately 33.333. A different exact supplied mean at that condition would not be reproduced by this unchanged pair S = 80 and τ = 0.005.

If r were unknown, the equations would contain an additional unknown and the supplied pair would not establish this unique two-parameter result. The useful answer is the conditional solution plus its boundary. It should not be presented as a general method for deriving instrument parameters from two readings whose input relationship was never independently established.

Contents · Previous chapter · Next chapter

The repair route begins with one forward response and one inverse family. Stabilise units, reciprocals and the distinction between input and recorded means. A learner should see why one equation leaves alternatives before trying to remove an unknown with a second condition.

The stabilisation route mixes compatible, linear-boundary and incompatible pairs. Require a forward check and a domain statement for each. Do not label every question with the formula to use; the learner must identify whether the supplied information actually matches the model.

The extension route introduces a third condition, uncertainty in r, near-equal readings and an unscaled background term. Ask which change affects identifiability and which affects validation. These are different questions even when both lead to a request for more evidence.

Keep a record containing raw observations, time basis, relative-input definition, response model, parameter solution, checks and remaining limitations. The useful output is an explanation another reader can reconstruct. No radiation equipment is needed for these educational calculations. Appropriate supplied datasets and conceptual models allow the learner to investigate the reasoning while respecting the professional scope of actual X-ray work.

Contents · Previous chapter · Next chapter

Can two readings always determine a recovery time?

No. The result depends on the supplied input ratio and response assumptions, and the pair must lie in the permitted domain. Equal readings or an incompatible ratio can prevent a finite physical solution.

Does a half input mean a half display?

Only in the linear boundary of this model. With positive recovery loss, the reduced input has a different loss fraction, so its displayed ratio is greater than the pre-loss ratio while remaining below one.

Does fitting two points validate the detector model?

It checks compatibility and algebra for those points. An independently supplied third condition tests another prediction. Actual validation needs suitable scope, uncertainty and physical evidence.

Can I assume a material or control setting gives the required ratio?

Not from its label alone. The known ratio must concern the eligible pre-dead-time event rates used in the model. Other changes in the measurement chain can affect that relationship.

What if a background does not scale with the source?

The main two-unknown shortcut no longer describes the same input model. Write the additional component explicitly and identify what evidence is needed rather than applying the earlier formulas unchanged.

Contents · Previous chapter · Next chapter

Return to X-ray Science for the full measurement chain and X-ray Spectroscopy for method and rate definitions. For interpretation limits, read one transmission reading and material identity.

For the separate known-parameter problem, use dead-time correction before background subtraction. For statistical context, use particle counters and counting uncertainty. Those articles supply different parts of the evidence chain rather than interchangeable correction recipes.

The specialist response-model reference is Chen, Lai and Li’s photon-counting detector primer, checked on 22 October 2026. The relative-input inverse, parameter pairs, domain comparisons and independent learner exercise here are original teaching developments from the explicitly supplied model. They are not experimental calibration results from that paper.

Continue into learning that transfers into life under How X Works. State what is known independently, solve only the model the record supports, and test the answer against a condition that did not merely define it.

Contents · Previous chapter · Return to X-ray Science