Circadian Science · LED Physics · Spectral Engineering
The Hidden Blue Spike
in "Warm" LEDs
There is no such thing as a truly blue-free phosphor-converted white LED. Every warm-white bulb — every 2700K, 3000K, and 4000K LED ever made — is built on a blue GaN semiconductor pump that emits intensely in the 440–460nm range. The warm colour comes from a phosphor that converts most of that blue into longer wavelengths. "Most" is not "all." The remainder — the unconverted blue pump residual — sits directly at the peak of the melanopsin action spectrum and is completely invisible in any CCT or CRI rating. This page explains the physics, shows you how to find it in a datasheet, and tells you exactly why it makes every warm-white LED inadequate for night-safe environments.
How White LEDs Actually Work — And Why They All Start Blue
Before you can understand why warm LEDs have a blue spike, you need to understand a fundamental fact about LED manufacturing: there is no such thing as a white LED semiconductor. White light from an LED is always produced indirectly — by converting some or all of the primary blue emission through a phosphor layer.
An LED — Light Emitting Diode — produces light through electroluminescence: electrons recombine with electron holes in a semiconductor junction, releasing energy as photons. The wavelength of those photons is determined entirely by the bandgap energy of the semiconductor material. You cannot change the emission wavelength by adjusting the circuit — it is locked in by the physics of the material.
The dominant material for high-efficiency, high-brightness LEDs is Gallium Nitride (GaN) and its related alloys (InGaN — Indium Gallium Nitride). GaN-based LEDs naturally emit in the blue-to-near-UV range, peaking between approximately 440nm and 470nm depending on the Indium content of the InGaN active layer. This is why virtually all efficient white LEDs are built on a blue foundation — GaN is the most efficient semiconductor available for high-brightness LEDs, and it naturally emits blue.
To make white light from a blue LED, manufacturers coat the LED chip with a phosphor layer — typically a cerium-doped yttrium aluminium garnet (Ce:YAG) or similar compound. When the blue photons from the GaN chip hit the phosphor, some are absorbed and re-emitted at longer wavelengths through a quantum-mechanical process called down-conversion (Stokes shift). By tuning the phosphor composition and thickness, manufacturers can shift the colour appearance of the output from cool-white (most blue passes through, less phosphor conversion) to warm-white (more blue absorbed, more phosphor emission at orange-red wavelengths).
The critical physics: phosphor conversion is never 100% efficient. Some fraction of the blue pump photons always passes through the phosphor layer unconverted. This fraction reaches the output as the residual blue spike visible in every white LED's spectral power distribution. It cannot be designed away within the phosphor-conversion architecture — it can only be reduced by increasing phosphor thickness or density, which comes at the cost of reduced luminous efficacy and phosphor heating.
Phosphor conversion efficiency in production Ce:YAG phosphors typically ranges from approximately 80% to 92% of incident blue photons at room temperature. This means that between 8% and 20% of the blue pump photons reach the output unconverted — as the blue spike. Higher phosphor density pushes the number lower but introduces other problems: more self-heating in the phosphor layer (which degrades it faster), higher light absorption (lower lumens), and more colour variation with temperature. There is no free lunch. The blue spike is the fundamental cost of the PC-LED architecture.
The Blue Spike: What It Is, What It Looks Like, Why It's Biologically Significant
On a spectral power distribution (SPD) plot of any white phosphor-converted LED, the blue spike appears as a sharply defined narrow peak near 440–460nm, separated by a shallow valley from the broader phosphor emission hump that creates the warm appearance. It looks small. Its circadian impact is disproportionately large.
Looking at a 2700K warm-white LED SPD, most of the visible emission is a broad, gently curved hump spanning roughly 500–700nm — the phosphor emission that creates the familiar warm golden-white appearance. But at around 440–460nm there is a narrow, asymmetric spike that rises sharply and falls sharply, separate from the phosphor band. This is the unconverted blue pump.
It looks small on the chart relative to the phosphor hump. But appearances deceive for two reasons. First, SPD plots are typically displayed on a relative scale normalised to the peak — the phosphor hump at its peak sets the 100% mark, and the blue spike appears as perhaps 20–40% of that height. In absolute photon terms, the spike still represents 8–20% of the total photon output at peak pump wavelength. Second, and critically, the sc(λ) melanopsin sensitivity function weights the spike region 20–100× more heavily than the phosphor hump region.
At 450nm — near the centre of the blue spike — sc(λ) is approximately 0.87 (87% of peak sensitivity). At 580nm — in the warm orange part of the phosphor hump — sc(λ) is approximately 0.01 (1% of peak sensitivity). The phosphor hump that creates all the warmth and visual light output contributes barely anything to melanopsin stimulation. The blue spike that represents a small visual fraction contributes almost everything to it.
This is the core asymmetry that makes the warm-LED circadian problem structurally unsolvable within the PC-LED architecture: the circadian-active portion of the output (the spike) and the visually dominant portion (the phosphor hump) are spectrally separate and cannot be adjusted independently. You cannot reduce the spike without reducing the pump drive current — which reduces all output proportionally. The spike's M/P ratio (its circadian impact per unit of visual light) is fixed by the LED physics, not by the colour temperature setting.
In a 2700K warm-white LED, the phosphor hump spanning 500–700nm contains roughly 80–90% of the total photon output — it's responsible for essentially all the visual light and all the warmth. Its circadian contribution per photon is tiny. The blue spike below 500nm contains perhaps 5–15% of total photons but contributes a disproportionate share of the melanopic stimulation because sc(λ) is so sensitive to that wavelength range. The result: a source that looks warm and gentle creates a circadian signal far larger than its appearance suggests, and no amount of phosphor adjustment can decouple these two components without abandoning the PC-LED architecture entirely.
How to Find the Spike in a Datasheet — What You're Looking For and Where to Find It
Not all manufacturer datasheets make the blue spike easy to see — some display SPD plots with Y-axis scales that make the spike look negligible. Here is exactly what to look for, what questions to ask, and what to do when a manufacturer won't provide an SPD.
What to request: Ask the manufacturer or supplier for the "Spectral Power Distribution" (SPD) or "Relative Spectral Power" (RSP) plot for the specific product at its rated drive current. This is a graph with wavelength in nm on the x-axis (typically 380–780nm) and relative power on the y-axis.
What to look for: In the 420–490nm region, look for a distinct narrow peak that rises and falls sharply. This is the blue spike. On a well-scaled chart it should be clearly visible as a separate feature from the broader phosphor emission. Some manufacturers use compressed y-axis scales or begin the wavelength axis at 400nm or higher — both of which can visually minimise the spike.
The right question to ask: "What percentage of the total radiant output falls below 500nm?" If the answer is greater than about 3%, the source will not meet the LumeCircadian M/P ≤ 0.02 specification. A product where the manufacturer cannot or will not answer this question should not be specified for a night-safe application.
Normalisation trap: Some SPD plots normalise to the phosphor peak (set to 100%), making the blue spike appear smaller relative to the full chart. Others normalise to the blue spike peak (set to 100%), which makes the phosphor hump appear as a lower broad feature. Both representations are accurate, but they create different visual impressions of the spike's relative importance. Always look at the actual wavelength range and the absolute spike amplitude relative to the phosphor hump — not just whether the spike looks "small."
Red flag — no SPD available: If a supplier markets an LED as "amber" or "warm night light" without publishing a spectral power distribution, treat it as disqualified until the SPD is obtained. Colour temperature, warm appearance, and even marketing images cannot substitute for actual spectral data when specifying for night-safe environments.
Misleading — spike underweighted visually
More accurate — spike correctly emphasised
"What is the percentage of total radiant output below 500nm?" This single number — which should be calculable from any SPD — immediately tells you the circadian risk level of the source. For a properly specified InGaAlP 590nm+ narrow-band amber emitter, the answer is close to zero. For a 2700K warm-white LED, the answer is typically 8–18%. Any supplier who cannot answer this question, or who deflects to CCT or CRI numbers, has not characterised their product for circadian applications.
Why CCT and CRI Completely Hide the Blue Spike
The two most common LED quality metrics — Colour Correlated Temperature (CCT, the K number) and Colour Rendering Index (CRI) — are both designed to describe visual appearance. Neither was designed to detect the blue pump spike, and neither does.
What CCT measures: Correlated Colour Temperature describes the chromaticity — the hue — of a light source on a scale from warm (red-amber, low K) to cool (blue-white, high K). It is calculated from the x,y chromaticity coordinates of the source on the CIE 1931 colour space. Two very different SPDs can produce the same chromaticity coordinates — and therefore the same CCT — if their spectral shapes average out to the same colour point.
This is exactly what happens with phosphor-converted warm LEDs: the blue spike and the broad phosphor hump combine to produce a chromaticity that maps to a warm colour temperature. The individual spike is invisible to the CCT calculation because CCT is computed from the aggregate colour, not the spectral detail. You cannot tell from the CCT whether a 2700K source has an 8% blue spike or a 20% one — both look identical to the CCT calculation.
What CRI measures: The Colour Rendering Index (Ra) measures how accurately a source renders the colour of test samples compared to a reference illuminant. It uses eight defined colour samples and computes average colour shift. The reference illuminant for warm sources (below ~5000K) is a Planckian (blackbody) radiator — which itself has no blue spike. A phosphor-converted LED can achieve a high CRI (>90) while having a substantial blue spike, because CRI tests the colour rendering across the visible range, not the shape of the SPD at any specific wavelength.
The bottom line: A 2700K, CRI 95 LED bulb tells you it produces a pleasant warm colour that renders colours accurately. It tells you nothing about its blue pump residual, its melanopic EDI, or its M/P ratio. A source with excellent CCT and CRI can simultaneously have a circadian impact 20× higher than an amber LED at the same lux level. The metrics are measuring visual quality — not circadian safety. These are not the same thing.
| Metric | What It Measures | Reveals Blue Spike? | Reveals Melanopic EDI? |
|---|---|---|---|
| CCT (Kelvin) | Visual colour appearance (warm/cool) | No | No |
| CRI (Ra) | Average colour rendering vs. reference | No | No |
| TM-30 Rf / Rg | Fidelity and gamut of colour rendering | No | No |
| Lux (photopic) | V(λ)-weighted illuminance | No | No |
| % output below 500nm | Short-wavelength spectral fraction | ✓ Yes | Partial |
| M/P Ratio (from SPD) | Melanopic-to-photopic ratio | ✓ Indirectly | ✓ Yes |
| Melanopic EDI (CIE S 026) | ipRGC-weighted circadian stimulation | ✓ Fully | ✓ Fully |
| Full SPD plot | Power at every wavelength | ✓ Directly visible | ✓ Calculable |
This metric gap is why the lighting advice "use 2700K warm bulbs at night" — dispensed by everyone from sleep clinics to smart home marketing — is not wrong, exactly, but dangerously incomplete. 2700K is better than 5000K for night use. But "better" is not the same as "safe." The 2700K rating tells you nothing about whether the source's M/P ratio is 0.40 or 0.52 — a difference that, at bedroom illuminance, means delivering 20 vs. 26 melanopic EDI lux. Both are above the clinical melatonin suppression threshold. Learn more in the clinical thresholds section.
Why Dimming Does Not Remove the Blue Spike — The Proportional Problem
The most commonly recommended nighttime lighting strategy — "use warm bulbs and keep them dim" — reduces melanopic EDI proportionally but never eliminates the M/P ratio problem. Dimming a phosphor-converted LED reduces everything equally. The spike shrinks, but it remains present at exactly the same fraction of total output.
Here is the core physics: dimming a phosphor-converted LED — whether by reducing drive current, applying PWM dimming, or using a wall dimmer — reduces the output of both the blue spike and the phosphor hump proportionally. If the spike is 15% of total photon output at full brightness, it is still 15% of total photon output at 10% brightness. The M/P ratio is a property of the spectral power distribution shape, not the absolute intensity level.
When you dim a 2700K LED from 100 lux to 20 lux, its melanopic EDI drops from ~45 mel. lux to ~9 mel. lux. That's genuine improvement — you've moved from well above the melatonin suppression threshold to approximately at the threshold for the general population. But you haven't solved the problem; you've changed its magnitude. At 20 lux, you're still stimulating melanopsin at a level that measurably affects sensitive individuals and infants.
The M/P ratio of the source is unchanged: it was 0.45 at full brightness, and it is still 0.45 at 10% brightness. To achieve ≤2 melanopic EDI lux at 20 photopic lux, you would need an M/P ratio of ≤0.10 — achieved only by candle flame (approximately) or by sources specifically designed with near-zero blue output below 500nm.
There is a secondary problem with dimming that most guides overlook: the typical dimming architecture for consumer LED products is PWM (pulse-width modulation), which introduces high-frequency flicker. Dimming a warm LED at night simultaneously fails on the circadian axis (spike still present) and introduces the neurological hazard covered in detail on the Flicker & Neuro page. Both hazards compound.
To get a 2700K warm-white LED below 2 melanopic EDI lux — the LumeCircadian nighttime target — you would need to reduce photopic illuminance to approximately 4.4 lux (2 ÷ 0.45 = 4.4). At 4.4 photopic lux, most adults cannot perform basic care tasks safely or comfortably in low-contrast environments. The 590nm amber source achieves 2 melanopic EDI lux at 100 photopic lux — more than enough light for any nighttime care task. Dimming the wrong source is not equivalent to using the right source.
How the Spike Varies Across CCT Values — Warmer Is Better, But Not Good Enough
Cooler-coloured LEDs have more prominent blue spikes than warmer ones — that much is intuitive. What's less obvious is the quantitative relationship: how much does the spike actually shrink as you move from 6500K to 2700K, and at what point does it become negligible? The answer is: never, within the phosphor-converted architecture.
| CCT | Typical % output <500nm | Spike peak wavelength | M/P Ratio (typical range) | Mel. EDI at 100 lux | vs. 10 mel. lux threshold |
|---|---|---|---|---|---|
| 6500K (blue-enriched) | 30–45% | ~455nm | 1.0–1.1 | ~100–110 lux | 10–11× above |
| 5000K (cool white) | 22–32% | ~450nm | 0.85–0.92 | ~85–92 lux | 8.5–9.2× above |
| 4000K (neutral white) | 15–22% | ~450nm | 0.60–0.70 | ~60–70 lux | 6–7× above |
| 3000K (warm white) | 10–16% | ~448nm | 0.48–0.58 | ~48–58 lux | 4.8–5.8× above |
| 2700K (warm white) | 8–14% | ~445nm | 0.40–0.50 | ~40–50 lux | 4–5× above |
| 2200K (high-CRI filament) | 3–8% | ~440nm | 0.18–0.24 | ~18–24 lux | 1.8–2.4× above |
| Candle (~1800K) | <3% | Minimal | 0.08–0.12 | ~8–12 lux | Near threshold |
| 590nm+ InGaAlP amber | ~0% | None | ≤0.02 | ≤2 lux | 5× below threshold |
Moving from 6500K to 2700K reduces the blue spike's fractional contribution from ~35% to ~11% of output. That is a meaningful improvement — the M/P ratio drops from ~1.05 to ~0.45. But 0.45 is still 4–5× above the 10 melanopic EDI lux clinical threshold at 100 photopic lux. The curve is flattening: going from 6500K to 2700K was a large gain. Going from 2700K to something genuinely night-safe requires stepping outside the phosphor-converted architecture entirely. The architecture itself is the ceiling. You cannot get below M/P ≈ 0.08 with any commercially realistic phosphor-converted warm-white LED.
The Biological Impact: Weighting the Spike Through sc(λ)
The blue spike does not contribute to melanopic EDI in proportion to its photon count — it contributes in proportion to its photon count multiplied by the melanopsin sensitivity at each wavelength. Because the spike falls at 440–460nm and melanopsin peaks at 480nm, the spike is weighted at 60–90% of maximum sensitivity. The result: a visually minor spectral feature carries disproportionate circadian weight.
To understand the disproportionate impact, consider a concrete calculation for a representative 2700K warm-white LED. Suppose the source emits 12% of its total photons below 500nm (the spike region) and 88% above 500nm (the phosphor hump region). The melanopsin sensitivity values sc(λ) at the relevant wavelengths are:
At 450nm (spike centre): sc(λ) ≈ 0.87 — 87% of peak melanopsin sensitivity
At 480nm (melanopsin peak): sc(λ) = 1.00 — maximum sensitivity
At 550nm (green, phosphor hump): sc(λ) ≈ 0.02 — 2% of peak
At 600nm (orange, phosphor hump): sc(λ) ≈ 0.003 — 0.3% of peak
At 640nm (red-orange, phosphor hump): sc(λ) ≈ 0.0005 — 0.05% of peak
When you multiply each region's photon fraction by its sc(λ) weight, the result is stark: the 12% of photons in the spike region contribute roughly 70–80% of the total melanopic signal. The 88% of photons in the phosphor hump contribute only 20–30%, and most of that comes from the 500–550nm green edge of the hump where sc(λ) is still non-trivial.
This is not a rounding effect or a minor correction. It is the fundamental reason why a visually warm, golden-white LED — which appears comfortably similar to firelight — carries a melanopic impact 4–5× larger than its visual character suggests. The visual system is insensitive to the spike. The circadian system is maximally sensitive to it.
For parents concerned about infant night environments, the consequence is even more significant: as documented on the Pediatric Shield page, infant crystalline lenses transmit approximately 90% of 450nm light to the retina (vs. ~60–65% for young adults). The spike that reaches a young adult retina at 87% melanopic weight reaches an infant retina at approximately 130% relative melanopic impact — because both the lens transmission and the melanopsin sensitivity are near-maximum at that wavelength.
If you could somehow remove just the blue spike from a 2700K warm-white LED — leaving all the phosphor hump emission intact — the resulting source would still look nearly identical (only slightly warmer) but its M/P ratio would drop from ~0.45 to approximately 0.05–0.10. The visual appearance is almost unchanged. The circadian impact is reduced by 75–90%. The spike is not incidental — it is almost the entire source of the circadian problem with warm-white LEDs. And you cannot remove it without removing the GaN pump that produces all the light.
The Only Real Solution: Spectral Architecture, Not Spectral Filtering
The phosphor-converted LED is not the only LED technology. There is a class of LEDs — Indium Gallium Aluminium Phosphide (InGaAlP) direct-emission devices — that produces light in the amber, orange, and red ranges without any blue pump whatsoever. This is the technology behind the LumeCircadian night specification.
InGaAlP LEDs are a different semiconductor family from GaN-based devices. Instead of a blue pump with phosphor conversion, they are direct-emission devices — the semiconductor bandgap is tuned to emit directly in the amber-to-red range by varying the Indium/Gallium/Aluminium ratio in the compound. There is no blue pump. There is no phosphor. There is no down-conversion. And therefore, there is no blue spike.
The emission from an InGaAlP device is a near-Gaussian spectral band centred at the design wavelength — for circadian applications, typically 590–615nm. Below 540nm, the spectral output of a well-specified InGaAlP amber LED is essentially zero. The sc(λ) values in that region are negligible (below 0.005 for any wavelength ≥590nm), and the resulting M/P ratio falls at or below 0.02.
The visual tradeoff: An InGaAlP amber emitter does not look like white light. At 590–610nm, it appears a saturated amber-orange — similar to amber street lights or a sodium vapour lamp, but cleaner and more controllable. This is not a "warm white" appearance. For nighttime care tasks — reading a baby's face, changing a nappy, navigating to the bathroom — it provides more than adequate scotopic (rod-based) vision at 10–20 lux. Human rod cells peak near 505nm and remain sensitive through 640nm, so scotopic acuity under amber light is functional even though the source looks monochromatic orange.
What InGaAlP amber is not: It is not a substitute for daytime lighting. A complete HCL (human-centric lighting) system uses the InGaAlP amber source exclusively for nighttime operation and a high-quality white LED (5000–6500K, CRI ≥ 90) for daytime use. The two channels are switched at sunset. This dual-channel approach is documented in detail on the HCL Retrofit Dual-Channel Wiring page.
The InGaAlP amber emitter does not eliminate the blue spike by filtering it. It eliminates it by not producing it in the first place. No GaN pump. No phosphor. No down-conversion inefficiency. No unconverted blue residual. The spectral output below 550nm is near-zero because the semiconductor physics dictate that no photons are generated there — not because anything has been done to remove them after the fact. This is the distinction between spectral architecture and spectral filtering — and it is the distinction that makes the difference between M/P ≤ 0.02 and M/P ≥ 0.05.
A common alternative proposed for night-safe lighting is adding a blue-blocking filter — an amber or orange lens — to a standard white LED fixture. This approach can reduce the blue spike output, but it does not eliminate it, and it introduces significant efficiency losses. A filter that blocks 95% of light below 500nm still passes 5% — which for a high-output fixture may still deliver meaningful melanopic stimulation. More importantly, the filter absorbs energy as heat, reducing fixture lifespan, and the filtering efficiency degrades over time with UV exposure. The fundamental advantage of InGaAlP direct-emission architecture is that zero blue output is an inherent material property, not a filtering outcome. It degrades with the emitter — on a 50,000-hour timescale — not with a separate component on an unknown degradation curve. See the emitter selection methodology for complete qualification criteria.