A filter described as “850 nm,” “550 nm longpass,” or “10 nm bandpass” is not fully specified by those nominal numbers alone.

Real optical filters are manufactured and measured against an allowable range. A nominal 850 nm bandpass filter, for example, may require its actual center wavelength to remain within a defined wavelength interval, while its bandwidth, transmission and blocking characteristics are controlled by separate limits.

This is the purpose of optical filter spectral tolerance.

Spectral tolerance defines how much a measured spectral characteristic may differ from its nominal or target value while the filter is still considered acceptable. It is not necessarily one number. Depending on the filter, separate tolerances may apply to center wavelength (CWL), FWHM, cut-on or cut-off wavelength, transmission, blocking, optical density and other spectral features. GIAI’s current optical-filter specification framework likewise treats CWL, FWHM, transmission, blocking, OD and AOI as separate engineering parameters rather than one universal filter rating.

Spectral Tolerance Is Not the Same as Nominal Wavelength

Suppose a drawing states:

Center wavelength: 850 nm

That tells the manufacturer where the passband is intended to be located, but it does not tell the manufacturer how far the actual production result is allowed to move.

A controlled specification might instead state:

CWL = 850 ± 2 nm

In this illustrative example, a measured CWL between 848 nm and 852 nm would satisfy the center-wavelength requirement.

But this still does not define the entire filter.

The same filter might also need a separate bandwidth specification, such as:

FWHM = 10 ± 2 nm

Commercial interference-filter specifications commonly separate center-wavelength tolerance from bandwidth tolerance rather than combining them into a single “spectral accuracy” figure. Published bandpass-filter examples, for instance, specify CWL tolerances and FWHM tolerances independently.

This distinction becomes increasingly important as the required passband becomes narrower.

---

1. Center Wavelength Tolerance

For a bandpass filter, the center wavelength, or CWL, identifies the spectral position of the passband.

For a symmetrical passband, it is often closely related to the midpoint between the two half-power wavelengths. The exact definition, however, should be agreed in the specification rather than assumed from the product name.

A requirement might be written as:

CWL = 532 nm ± 1 nm

or

CWL = 850 nm ± 3 nm

The tighter the allowed tolerance relative to the bandwidth, the more important manufacturing control and spectral verification become.

A ±2 nm CWL variation may be relatively unimportant for a 100 nm-wide broadband filter, but the same ±2 nm variation may consume a significant part of the usable passband of a 5 nm narrowband filter.

That is why wavelength tolerance should be selected according to the system requirement, not simply by choosing the smallest available number.

---

2. FWHM Tolerance

FWHM — full width at half maximum — describes the spectral width of a bandpass filter measured between the two wavelengths where transmission reaches half of the relevant peak level.

For example:

Nominal FWHM: 10 nm
Tolerance: ±2 nm

would allow an actual FWHM from 8 nm to 12 nm, provided the agreed measurement definition is satisfied.

FWHM tolerance matters because bandwidth affects both wanted signal throughput and unwanted spectral rejection.

A wider passband may collect more useful signal but can also admit more background radiation. A narrower passband may improve spectral discrimination but becomes more sensitive to wavelength positioning, angular distribution and other operating conditions.

Therefore, CWL tolerance and FWHM tolerance should normally be evaluated together.

---

3. Cut-On and Cut-Off Wavelength Tolerance

Longpass, shortpass, dichroic and some bandpass filters are often better specified by their spectral edges rather than only by a center wavelength.

A longpass specification, for example, may define a cut-on wavelength near 550 nm.

But there is an important question:

What exactly does “550 nm cut-on” mean?

Depending on the specification, the edge may be defined as:

• the wavelength at 50% transmission,
• a wavelength at another specified transmission level,
• the end of a blocking region,
• or the beginning of the required passband.

GIAI’s current technical guidance explicitly notes that filters with the same nominal cutoff can have different transition slopes and usable spectral behavior if the cutoff definition is not stated.

A clearer drawing might therefore specify:

50% cut-on wavelength: 550 ± 3 nm

rather than simply:

550 nm longpass filter

Commercial filter specifications also commonly give cut-on and cut-off tolerances independently, demonstrating why an edge wavelength should not be treated as an exact, dimensionless label.

---

4. Transmission Tolerance Is a Different Requirement

Spectral position is only part of filter performance.

A bandpass may be correctly centered and have the correct bandwidth but still fail because its transmission is too low.

Transmission requirements may be defined as:

• minimum peak transmission,
• minimum transmission at a particular wavelength,
• average transmission across a wavelength interval,
• or transmission uniformity across a specified band.

These definitions are not interchangeable.

For example:

Average transmission ≥ 90% from 845–855 nm

is a different acceptance criterion from:

Peak transmission ≥ 90%

A filter with a narrow 92% peak but significantly lower transmission through the rest of the passband could meet the second specification while failing the first.

The test definition therefore needs to match what the optical system actually requires.

---

5. Blocking and Optical Density Need Their Own Limits

A filter may have an excellent passband and still perform poorly in the system if unwanted wavelengths leak through.

Blocking should normally define both:

1. how much attenuation is required, and
2. over what wavelength range it is required.

Optical density is related to transmittance by:$$OD=-\log_{10}(T)$$

where $T$ is transmittance expressed as a decimal fraction.

An OD requirement without a wavelength range is incomplete.

For example:

OD ≥ 4 from 400–800 nm outside the defined passband

contains substantially more engineering information than simply:

OD4 filter

The relevant blocking range should reflect the actual source spectrum, detector response and unwanted optical signals in the system.

---

An Illustrative Spectral Tolerance Specification

The following example is for explanation only and is not a GIAI product specification.

Parameter	Illustrative Requirement
Center wavelength	850 ± 2 nm
FWHM	10 ± 2 nm
Peak transmission	≥ 90%
Blocking	OD ≥ 4 from 400–800 nm and 900–1100 nm
AOI	0°
Polarization	Unpolarized
Measurement condition	Defined by agreed inspection method

This specification describes several independent acceptance criteria.

A filter measuring:

• CWL = 851 nm
• FWHM = 11 nm
• Peak T = 92%

could satisfy those three requirements.

However, it could still fail the specification if out-of-band blocking does not meet the defined OD requirement.

This is why comparing only the peak wavelength of two spectral curves is not sufficient for filter acceptance.

---

Why Does the Spectrum Vary?

Spectral tolerance exists because the final spectrum depends on the physical filter structure, manufacturing process and measurement conditions.

For interference filters, wavelength-selective behavior is produced by multilayer thin-film structures. Small variations in optical thickness or refractive index can change the location and shape of the final spectral response.

But manufacturing variation is not the only cause.

The spectrum measured in an application can also differ from a normal-incidence laboratory curve because of angle of incidence, beam geometry, polarization, temperature and test configuration.

---

Angle of Incidence Can Move the Passband

Angle of incidence is especially important for interference filters.

As the incidence angle increases from normal incidence, interference-filter spectral features generally move toward shorter wavelengths. This is commonly called a blue shift.

A simplified relationship is:$$\lambda_\theta \approx \lambda_0 \sqrt{1-\frac{\sin^2\theta}{n_{\mathrm{eff}}^2}}$$

where:

• $\lambda_0$ is the wavelength at normal incidence,
• $\lambda_\theta$ is the shifted wavelength,
• $\theta$ is the incidence angle,
• $n_{\mathrm{eff}}$ is an effective refractive index representing the multilayer structure.

The expression is an engineering approximation; the exact response depends on the actual coating design. The general blue-shift behavior of interference filters with increasing AOI is well established.

This has an important specification consequence:

A wavelength tolerance without an AOI condition may be incomplete.

For example, a CWL tolerance measured at 0° should not automatically be interpreted as the filter’s CWL when it operates at 20° inside an optical system.

---

Do Not Ignore the Beam Cone

AOI is also not always one single ray angle.

In an imaging system or converging beam, different rays can reach the filter at different incidence angles. A wide angular distribution can therefore broaden, shift or reshape the observed spectral response.

For narrowband filters, the effect may be much more important than it is for a broad spectral filter.

The useful specification may therefore need to describe:

• chief-ray AOI,
• angular range,
• cone half-angle,
• numerical aperture,
• or the actual optical geometry.

This is one reason a spectral curve measured with a nearly collimated laboratory beam should not automatically be assumed to represent performance in every installed optical system.

---

Polarization Can Matter at Oblique Incidence

At normal incidence, polarization effects may be relatively small for many filter designs.

At larger AOI, however, the spectral response for s- and p-polarized light can diverge.

If the system uses a defined polarization state, or if the filter operates at a significant angle, the procurement specification should state whether spectral targets apply to:

• s-polarization,
• p-polarization,
• unpolarized light,
• or an agreed polarization average.

Otherwise two parties may measure the same filter under different polarization conditions and obtain different results without either measurement being inherently wrong.

---

Temperature May Also Affect the Spectrum

Temperature can change optical properties and physical layer dimensions, so some interference filters exhibit a measurable wavelength shift as temperature changes.

The magnitude is not universal. It depends on material system and filter design.

Published interference-filter products and research both demonstrate measurable temperature-dependent wavelength shifts, so applications with a wide operating-temperature range should define the relevant environmental condition rather than assume room-temperature behavior is sufficient.

---

Spectral Tolerance vs Measurement Uncertainty

Another important distinction is between:

the allowed product tolerance

and

the uncertainty or capability of the measurement method.

If a drawing requires CWL = 850 ± 1 nm, the inspection method must be capable of resolving the spectral feature well enough to make a meaningful pass/fail decision.

Other test details can also affect the measured result:

• spectrophotometer wavelength accuracy,
• spectral sampling interval,
• spectral resolution,
• beam divergence,
• sample orientation,
• AOI,
• polarization,
• baseline correction,
• and treatment of very low transmission when calculating OD.

For this reason, engineering and purchasing teams should avoid comparing curves from different laboratories without first checking whether the measurement conditions are comparable.

GIAI’s current quality-control framework follows this same principle: spectral verification for wavelength-selective coated components is defined against the individual component specification, with test conditions and acceptance limits tied to the agreed project requirements rather than one universal inspection rule.

---

How Tight Should Optical Filter Spectral Tolerance Be?

The answer should come from the optical system.

A tighter number is not automatically a better specification.

A useful tolerance should consider:

Source spectrum.
How stable is the LED, laser, lamp or other source wavelength?

Detector response.
How quickly does detector sensitivity change across the wavelength region?

Required signal-to-background ratio.
Does a small wavelength displacement significantly increase background?

Filter bandwidth.
A ±2 nm CWL tolerance means something very different for a 5 nm filter and a 100 nm filter.

AOI and angular distribution.
Will the filter operate at normal incidence or inside a converging or scanned beam?

Temperature range.
Is the system operated only near room temperature or over a wider environment?

Manufacturing margin.
Does the requested tolerance actually improve system performance enough to justify tighter process and inspection control?

The objective is not to request the smallest possible tolerance. It is to define a tolerance that protects system performance while remaining technically meaningful and manufacturable.

---

A Better Way to Write an Optical Filter Specification

Instead of sending:

Need an 850 nm filter with high transmission and OD4.

a more useful technical inquiry would define:

Bandpass filter
Target CWL and allowed tolerance
FWHM and bandwidth tolerance
Minimum or average passband transmission
Blocking wavelength range
Required OD within that range
Operating AOI and angular range
Polarization condition if relevant
Substrate
Filter dimensions and clear aperture
Operating environment
Inspection and reporting requirements
Prototype and expected production quantity

This provides both the coating engineer and the inspection team with a common acceptance basis.

GIAI’s custom-filter review process follows the same project-specific approach: wavelength, transmission or reflection, blocking, substrate, AOI, geometry and inspection requirements are reviewed together before the manufacturing and verification route is finalized. The project reference likewise requires custom optical work to be evaluated against the actual drawing, optical requirements, substrate, geometry, coating conditions and inspection criteria rather than assumed generic capability.

---

How Should Spectral Tolerance Be Verified?

For incoming inspection or supplier qualification, do not ask only whether the curve “looks similar” to a reference plot.

Create acceptance criteria that can be evaluated objectively.

Depending on the filter, this may include:

• CWL limits,
• FWHM limits,
• cut-on or cut-off wavelength limits,
• minimum transmission,
• average transmission,
• blocking or OD limits,
• specified wavelength intervals,
• AOI,
• polarization,
• and defined measurement conditions.

A typical spectral curve is useful for understanding expected behavior, but a curve alone should not replace the written acceptance specification.

This is particularly important in production because a typical or representative curve does not mean every unit will reproduce every plotted point exactly. The controlled specification should determine whether a filter passes or fails.

---

FAQ

Is spectral tolerance the same as wavelength tolerance?

Not always. Wavelength tolerance usually refers to the allowed displacement of a defined wavelength feature such as CWL or an edge wavelength. Spectral tolerance can be broader and may include bandwidth, transmission, blocking and other spectral acceptance limits.

Is ±2 nm a tight optical-filter tolerance?

It depends on the filter bandwidth, operating wavelength, optical system and measurement conditions. ±2 nm may be insignificant for one broadband filter and critical for another narrowband filter.

Should CWL tolerance be specified for a longpass filter?

Usually the more meaningful parameter for a longpass or shortpass filter is an explicitly defined edge wavelength such as the 50% cut-on or cut-off point rather than a CWL.

Does tilting an interference filter change its wavelength?

Yes. Increasing AOI generally shifts interference-filter features toward shorter wavelengths, although the exact shift depends on the coating design.

Can I specify only the target wavelength when requesting a custom filter?

For early discussion, yes, but it is normally insufficient for manufacturing acceptance. Bandwidth or edge definition, transmission, blocking, AOI, dimensions and other system-dependent requirements should be added as the design is finalized.