Key equations and principles
Quick reference of conversion factors for common wavelengths
In the table, ‘GHz/nm’ = c/λ₀², i.e. the frequency interval corresponding to 1 nm of bandwidth at that wavelength; ‘1 GHz → pm’ is its reciprocal, used to convert a frequency linewidth back to a wavelength linewidth. Values are calculated with c = 299 792 458 m/s.
| Laser wavelength λ₀ | Centre frequency ν₀ (THz) | 1 nm → GHz | 1 GHz → pm | Typical source |
|---|---|---|---|---|
| 405 nm | 740.228 | 1827.72 | 0.5471 | Blue-violet diode laser / maskless lithography |
| 488 nm | 614.329 | 1258.87 | 0.7944 | Ar⁺ laser / flow cytometry |
| 532 nm | 563.520 | 1059.25 | 0.9441 | Frequency-doubled Nd:YAG, green pointer |
| 633 nm | 473.606 | 748.19 | 1.3366 | He-Ne laser / interferometry |
| 780 nm | 384.349 | 492.76 | 2.0294 | Rb atomic clock / atomic physics |
| 808 nm | 371.030 | 459.20 | 2.1777 | Nd:YAG pumping |
| 915 nm | 327.642 | 358.08 | 2.7927 | Fibre laser pumping |
| 1030 nm | 291.061 | 282.58 | 3.5388 | Yb:YAG femtosecond laser |
| 1064 nm | 281.760 | 264.81 | 3.7763 | Nd:YAG / fibre laser |
| 1310 nm | 228.849 | 174.69 | 5.7243 | O-band optical communication (zero dispersion) |
| 1550 nm | 193.414 | 124.78 | 8.0139 | C-band optical communication / eye-safe |
| 2000 nm | 149.896 | 74.95 | 13.3426 | Mid-infrared pumping / medical |
| 10.6 μm | 28.282 | 2.67 | 374.79 | CO₂ laser / cutting |
Calculation examples
Example 1: 1064 nm Nd:YAG with a 1 nm wavelength bandwidth
- Band edges: λ₁ = 1063.5 nm, λ₂ = 1064.5 nm
- Δν = c·Δλ/(λ₁λ₂) = 2.998×10⁸ × 1×10⁻⁹ / (1.0635×10⁻⁶ × 1.0645×10⁻⁶) ≈ 264.81 GHz
- Approximation check: c·Δλ/λ₀² = 264.81 GHz, differing from the exact value by 0.00002% (the bandwidth is only 0.094% of the centre wavelength)
- Coherence length Lc = c/Δν ≈ 1.132 mm;Q = ν₀/Δν ≈ 1064
Example 2: 1550 nm WDM — how many nm does a 100 GHz channel spacing correspond to?
- ν₀ = c/λ₀ ≈ 193.41 THz,ν₁ = 193.4645 THz,ν₂ = 193.3645 THz
- λ₁ = c/ν₁ ≈ 1549.599 nm,λ₂ = c/ν₂ ≈ 1550.401 nm
- Δλ = λ₂ − λ₁ ≈ 0.8014 nm (the industry rule of thumb ‘100 GHz ≈ 0.8 nm’ comes from here)
Example 3: 800 nm Ti:sapphire — how short a pulse can a 10 nm bandwidth support?
- Δν = c·Δλ/(λ₁λ₂),λ₁ = 795 nm、λ₂ = 805 nm → Δν ≈ 4.6844 THz
- Approximation error 0.0039% (the bandwidth is already 1.25% of the centre wavelength, but the error is still negligible)
- Gaussian transform limit Δt = 0.441/Δν ≈ 94.1 fs; for sech² it is 0.315/Δν ≈ 67.2 fs
Example 4: 1550 nm DFB (distributed feedback) laser — what wavelength does a 1 MHz linewidth correspond to?
- Δλ = c·Δν/(ν₀² − Δν²/4) ≈ 8.014×10⁻¹⁵ m
- 8.0 fm cannot be read at all in the wavelength domain — this is exactly why narrow-linewidth lasers always specify linewidth in frequency
- Coherence length Lc = c/Δν = 299.79 m
Common misuses
Substituting λ and Δλ in the wrong places
The denominator of Δν = c·Δλ/λ₀² is thesquare of the centre wavelength, the numerator is thebandwidth. Solving in reverse for wavelength bandwidth gives Δλ = λ₀²·Δν/c. Putting the bandwidth in the denominator and the wavelength in the numerator is the most frequent calculation error, off by up to a factor of 10⁶.
Still using the approximation for broad bandwidths
The deviation of the approximation is (Δλ/2λ₀)².At a bandwidth of 10% of the wavelength the deviation is about 0.25%, and at 50% about 6.7%, a level often reached by supercontinuum sources, LEDs and femtosecond oscillator outputs. For such inputs, refer directly to the ‘exact value’ in the data table.
Confusing pulse duration, coherence time and 1/Δν
All three equal ‘constant/Δν’, but with different constants:Gaussian transform-limited pulse duration 0.441, Gaussian |g₁| coherence time 0.530, engineering convention 1, Lorentzian 0.318. Using 1/Δν as the pulse duration overestimates by about a factor of 2.3 (equivalent to overestimating the pulse width, Gaussian), enough to misjudge the order of magnitude in ultrafast experiments.
Frequently asked questions (FAQ)
Why is 1 nm equal to 265 GHz at 1064 nm but only 125 GHz at 1550 nm?
Because the denominator of Δν ≈ c·Δλ/λ₀² is the square of the wavelength. The longer the wavelength, the smaller the relative bandwidth for the same wavelength interval, and the smaller the frequency interval. The coefficient c/λ₀² is 264.8 GHz/nm at 1064 nm and 124.8 GHz/nm at 1550 nm, differing by exactly (1550/1064)² ≈ 2.12. So ‘how many GHz is 1 nm’ must always be stated with the wavelength; on its own it is meaningless.
Which is more fundamental, Δλ or Δν?
The frequency domain is more fundamental. Fourier transforms, photon energy, coherence time and transform-limited pulse duration all take frequency as the natural variable, and the relation between frequency bandwidth and lineshape is more direct (a Gaussian spectrum remains Gaussian in the frequency domain but becomes slightly asymmetric when mapped to wavelength). The advantage of the wavelength domain is that it matches the labelling conventions of spectrometers, coatings and filters. Neither is right or wrong, but for coherence or ultrafast pulse durations it is advisable to standardise on Δν.
When must the exact formula be used?
The criterion is the relative bandwidth Δλ/λ₀. When the bandwidth is below 1% of the centre wavelength the approximation error is about 0.0025% (2.5×10⁻⁵), completely negligible; at 10% it is about 0.25%, still acceptable for most engineering; above 20%–30% (supercontinuum, LED, broadband femtosecond) the exact formula should be used. The data table of this tool gives both values and the relative deviation, with an automatic warning above 5%.
Why does a band symmetric in the frequency domain become asymmetric in the wavelength domain?
Because λ = c/ν is a hyperbolic mapping, not a linear scaling. Equally spaced ticks in the frequency domain have different spacings everywhere after mapping to wavelength (e.g. 100 GHz channels equally spaced in frequency have unequal wavelength intervals). Strictly speaking, a band of ‘centre wavelength ±Δλ/2’ is not centred on ν₀ in the frequency domain — the short-wavelength side spans a larger frequency range; the ‘frequency-domain asymmetry’ in this tool’s data table is exactly this ratio, which is extremely close to 1 for narrow bands and can be ignored.
Should the coherence length use c/Δν or λ₀²/Δλ?
The two are equivalent for narrow bands: c/Δν = c/(cΔλ/λ₀²) = λ₀²/Δλ. The strict form is Lc = λ₁λ₂/Δλ, exactly consistent with c/Δν (since Δν = cΔλ/(λ₁λ₂)). The only difference is whether the Δλ²/4 term is neglected; for bandwidths below 1% of the wavelength the two differ by less than 10⁻⁴ in relative terms.
Why are there three forms of coherence time: 1/Δν, 0.530/Δν and 1/(πΔν)?
Because ‘coherence time’ has no unique definition — it depends on the lineshape and criterion. The engineering convention τc = 1/Δν is used for order-of-magnitude estimates; for a Gaussian spectrum, |g₁(τ)| falling to 1/e corresponds to 0.530/Δν; for a Lorentzian spectrum to 1/(πΔν) = 0.318/Δν. The differences between these three values arise from different definitions of lineshape and criterion, representing real physical differences, not calculation errors.
Where do the 0.441 and 0.315 for transform-limited pulse duration come from?
They come from the analytical result of the Fourier transform of the lineshape: for a spectrum with Gaussian intensity and FWHM Δν, the intensity envelope of its Fourier transform has FWHM 0.441/Δν; for a sech² spectrum it is 0.315/Δν; for a rectangular spectrum 0.886/Δν. The constant depends only on the lineshape, not on the centre wavelength. Real lasers are always wider due to chirp and dispersion, so these two values serve only as lower-limit references.
How does this tool differ from the site’s ‘wavelength/frequency/photon energy converter’?
That tool handlespoint-value conversion: 1064 nm ↔ 281.76 THz ↔ 1.165 eV, valid for any wavelength (wavelength/frequency/photon energy converter). This tool handlesinterval-width conversion: Δλ ↔ Δν. Bandwidth cannot simply copy the point-value ratio, because λ and ν are nonlinearly related — this is precisely why this tool is provided separately. Use the two together: first fix the centre with the point-value conversion, then convert the width with this tool.
How do I choose units? Which should I use: pm, nm, GHz, THz or cm⁻¹?
By convention: coatings, filters and spectrometers use nm or pm; narrow-linewidth lasers, WDM and coherent communications use GHz or MHz; spectroscopy (Raman, absorption) uses cm⁻¹; ultrafast work uses THz together with fs. This tool computes entirely in SI internally and automatically selects a suitable magnitude for output. cm⁻¹ and GHz are strictly linear (1 cm⁻¹ = 29.9792458 GHz), so you can multiply or divide directly without worrying about approximation.
Can it still calculate when the bandwidth exceeds 10% of the centre wavelength?
Yes, but the basis must be clear. Mathematically it holds as long as the band edges remain positive; physically, for broad bands the very definitions of ‘centre wavelength’ and ‘FWHM’ become ambiguous (the spectrum is strongly asymmetric), so it is more appropriate to give the full spectral curve rather than a single bandwidth value. This tool prompts you to use the exact value when the relative bandwidth exceeds 5% and recommends direct integration of the spectral data above about 30%.
How do I convert between FWHM, RMS and 1/e² bandwidth?
A definite relation exists only for Gaussian spectra: FWHM = 2.355σ (σ is the standard deviation, i.e. the RMS width). The 1/e full intensity width = 2√2σ ≈ 2.828σ, hence FWHM = 0.833 × (1/e full width); the 1/e² full intensity width commonly used in datasheets (where the intensity falls to 1/e² of the peak) = 4σ, hence FWHM = 0.589 × (1/e² full width), i.e. the 1/e² full width ≈ 1.699 × FWHM. A Lorentzian spectrum has no finite RMS width (the integral diverges) and cannot be converted this way. All inputs and outputs of this tool use the intensity full width at half maximum (FWHM) basis; mixing other bases will systematically overestimate or underestimate the result.


