Astronomy · Learning from more than light

What different detectors actually measure

Trace incident particles and gravitational-wave strain through interactions, sensor readouts and response models. Calculate selected-count expectations and a differential-arm example while keeping them distinct from observations.

Continue in the approved chapter order. Necessary vocabulary, quantities and assumptions are explained in the lesson.

Back to the chapter

After reading, you can

Explain the difference between a primary signal, raw readout and reconstructed event.
Apply a supplied two-channel response matrix with its units, migration and background.
Calculate the single-arm and differential responses in a declared weak-wave geometry.

The detector does not hand us the source itself

A catalogue can list an event time, energy estimate and reconstructed direction in a few columns. Those compact entries sit at the end of a chain: a primary signal interacts, sensors respond, electronics record a signal, an analysis selects events and a model connects the retained record to incident quantities. Each stage can introduce information loss and uncertainty.

An event is therefore defined by an analysis as well as by hardware. A pulse can come from a desired interaction, an atmospheric background or instrumental noise. A trigger decides which conditions deserve recording or rapid attention; it does not by itself establish an astronomical origin.

Calibration relates the readout to specified quantities under stated conditions. Reconstruction uses response and physical models to estimate properties such as deposited energy, incident direction or strain. Inferring the distant source then adds a source and propagation model. Accurate terminology helps us locate which part of the argument the data actually test.

The examples below are original supplied models. Their matrices, arm lengths, strain and count means are not measurements of any observatory. We use them to expose the reasoning that an actual experiment must justify.

Light in a neutrino detector belongs to the interaction chain

The IceCube instrumentation paper describes optical modules that record light from charged secondary particles produced in neutrino interactions. A suitable secondary emits Cherenkov radiation when its speed exceeds the phase speed of light in the medium. This does not require a particle to exceed the vacuum light speed c.

The light pattern and timing can support a track-like or shower-like event reconstruction. An observed energy deposit need not equal the entire incident neutrino energy: the interaction channel, escaping particles, detector containment and response matter. Flavour-sensitive interactions and propagation also prevent an event label from being a direct source-flavour inventory.

Charged cosmic rays can initiate cascades of secondary particles in the atmosphere. Detectors can sample particles at the ground or light produced along the cascade. The Particle Data Group review explains why reconstructing a primary energy or composition requires an interaction and shower model, rather than treating every local sensor signal as the primary particle itself.

These are complementary techniques, not a universal design shared by every neutrino or cosmic-ray detector. Their historical instrumentation descriptions supply the physical measurement examples here; no present detector layout, sensitivity, threshold or resolution is asserted.

A response matrix predicts selected events

For a hypothetical counting detector, let Φ1 and Φ2 be incident particle fluences in two true-energy bins, in particles m⁻², integrated over the same stated time interval. Fix the arrival direction, the spectrum shape within each bin and all response conditions. Label the two reconstructed channels A and B; these labels do not mean the true-energy bins have been measured perfectly.

Supply effective-area coefficients M_A1 = 0.4 m², M_A2 = 0.1 m², M_B1 = 0.2 m² and M_B2 = 0.3 m². The expected source counts are s_A = 0.4 m² × Φ1 + 0.1 m² × Φ2 and s_B = 0.2 m² × Φ1 + 0.3 m² × Φ2. Area times fluence gives a dimensionless expected event count.

In this particular construction, the geometric area is 1 m² and each coefficient is that area times a disjoint channel-selection probability. The column sums are 0.6 m² and 0.4 m², corresponding to total detection probabilities 0.6 and 0.4. A true-bin-1 particle may be assigned to either channel or missed. Effective area has a more general meaning in real instruments and need not always equal a physical collecting area times this simple probability.

Finally stipulate known mean background counts b_A = 2 and b_B = 3 in the same interval. Expected total counts are μ_A = s_A + 2 and μ_B = s_B + 3. These are supplied means, not measured background samples or guaranteed integer outcomes. A probability model for actual observations would be an additional part of an inference.

Original hypothetical effective-area response. Rows are reconstructed channels; columns are true-energy bins. Each entry has units m².
Reconstructed channel	True-energy bin 1	True-energy bin 2
A	0.4	0.1
B	0.2	0.3

Selection and migration change the apparent spectrum

For Φ1 = 30 and Φ2 = 10 particles m⁻², source expectations are s_A = 12 + 1 = 13 and s_B = 6 + 3 = 9. Adding background gives μ_A = 15 and μ_B = 12. The incident total is 40 particles m⁻², but the selected-source total is 22 expected events in this supplied instrument model.

Swapping the incident bin fluences to 10 and 30 keeps the total incident fluence at 40 particles m⁻². The source expectations become 7 and 11, with total means 9 and 14 after background. The selected-source total falls to 18 because the second true-energy bin has lower total detection probability.

A third input of 20 particles m⁻² in each true bin gives source means 10 and 10. Equal channel means do not imply perfect one-to-one energy assignment; both reconstructed channels contain contributions from both true bins. The full table retains all three supplied comparisons.

Even when a small matrix has an algebraic inverse, observed counts fluctuate and the response and backgrounds have uncertainty. In larger problems, response columns can be similar or some incident signals can have nearly zero sensitivity. Solving a linear system alone does not establish a unique, stable or physically adequate incident spectrum.

All three original incident-fluence comparisons, with b_A = 2 and b_B = 3 known mean counts. Each input has total fluence 40 particles m⁻². No random observed sample is supplied.
Φ1 (particles m⁻²)	Φ2 (particles m⁻²)	s_A	s_B	μ_A	μ_B
30	10	13	9	15	12
10	30	7	11	9	14
20	20	10	10	12	13

Each arm changes by half the specified plus strain in this geometry

A gravitational-wave interferometer is not described by the particle-counting matrix above. Its optical readout and feedback signals are calibrated into a strain time series. The LIGO calibration paper shows how sensing and actuation models enter that reconstruction, including their frequency response and uncertainty.

Construct a separate ideal example with two unperturbed equal arms of length L = 20 m along x and y. A plus-polarized weak plane wave travels normal to their plane, with the arms aligned to its plus axes and no cross polarization. Use the freely falling test-mass and long-wavelength approximation at one stipulated instant, without a feedback system in this toy.

To first order, the transverse metric response gives δL_x/L = +h_plus/2 and δL_y/L = −h_plus/2. One way to see the half factor is to expand √(1 ± h_plus) ≈ 1 ± h_plus/2 for |h_plus| much smaller than one. The sensed differential strain is h_det = (δL_x − δL_y)/L = h_plus for this particular orientation.

Choose h_plus = 2 × 10⁻⁹. Then δL_x = +2 × 10⁻⁸ m = +20 nm and δL_y = −20 nm. Their difference is 40 nm, and 40 × 10⁻⁹ m divided by 20 m returns h_det = 2 × 10⁻⁹. Calling each individual arm change h_plus L would incorrectly double the differential response.

A calibrated quantity still carries a model and uncertainty

The ideal arm example isolates geometry. Real interferometers have optical cavities, suspended mirrors, control loops, calibrated amplitude and phase, and direction- and polarization-dependent response. A measured strain channel does not independently recover every incoming polarization or source parameter.

For a linear time-invariant instrument in sinusoidal steady state, a transfer function H(ω) = Y(ω)/X(ω) describes the gain and phase between input and output phasors. Here ω is angular frequency in rad s⁻¹, and H has output-unit divided by input-unit. That model requires a nonzero input phasor and a calibrated range; it is not a universal response law for every detector.

Likewise, a reconstructed arrival time may depend on pulse shape, filtering, threshold and clock calibration. An incident energy estimate may depend on which energy escaped the instrument and on the assumed interaction. Report those response conditions alongside the event estimate instead of giving the estimate the authority of an unprocessed fact.

The next lesson uses calibrated arrival times and detector locations to constrain a sky direction. It will preserve the distinction established here: the raw signal, its reconstruction and the proposed astronomical interpretation are related stages with separate assumptions.

Worked examples

Follow both true bins into one channel

For incident fluences 30 and 10 particles m⁻², why is channel A’s total mean 15 rather than 30?

The supplied response assigns 0.4 m² × 30 = 12 and 0.1 m² × 10 = 1 expected source events to A.
The source mean in A is 13.
Adding its supplied background mean of 2 gives total mean 15.

Channel A is a selected reconstructed mixture of the two incident bins, with background. Its mean is not a direct incident-bin count.

Check the half factor and the subtraction

What changes and differential strain follow from L = 20 m and h_plus = 2 × 10⁻⁹ in the declared ideal geometry?

Each arm has fractional change magnitude h_plus/2 = 10⁻⁹.
The arm changes are +20 nm and −20 nm, giving a difference of 40 nm.
Divide the difference by 20 m to obtain 2 × 10⁻⁹.

The differential strain equals h_plus here, while each arm has half that fractional change. These are stipulated teaching values, not detector performance.

Try the reasoning

In the supplied response matrix, what do the rows and columns describe?
Show the answer and reasoning
Rows are reconstructed channels; columns are true-energy bins.
Each incident true bin can contribute to both recorded channels, with some particles missed.
Must equal total incident fluences give equal selected-source count totals in this detector?
Show the answer and reasoning
No.
The true bins have different detection probabilities; the supplied 30/10 and 10/30 mixtures give 22 and 18 expected source events.
Does a model mean μ_A = 15 guarantee that an actual exposure records exactly 15 events?
Show the answer and reasoning
No.
A mean is an expectation. A statistical observation model and its fluctuations must be specified separately.
In the ideal plus-aligned weak-wave example, is one arm’s fractional change h_plus or h_plus/2?
Show the answer and reasoning
Its magnitude is h_plus/2.
The two arms change oppositely, so their difference divided by L is h_plus in that geometry.

What to carry forward

A reconstructed event connects a raw record to an incident quantity through response, selection and calibration.
A counting response maps true-bin fluences into channel expectations; those expectations are not observations.
The ideal gravitational-wave differential response and a real calibrated strain time series require their own geometry and instrument models.

References

Teaching resources and research supporting this lesson’s explanations and examples.