Color Maintenance of LEDs

in Laboratory and Field

Applications

September 2013

Prepared for:
Solid-State Lighting Program
Building Technologies Office
Office of Energy Efficiency and
Renewable Energy
U.S. Department of Energy
Color Maintenance of LEDs in Laboratory and Field Applications

Michael Royer,1 Ralph Tuttle,2 Scott Rosenfeld,3 Naomi Miller1

1
Pacific Northwest National Laboratory, Portland, OR
2
Cree, Inc., Durham, NC
3
Smithsonian Institution, Washington, DC

September 2013
PNNL-22759

Abstract
To date, consideration for parametric failure of LED products has largely been focused on lumen
maintenance. However, color shift is a cause of early failure for some products, and is especially
important to consider in certain applications, such as museum lighting, where visual appearance is
critical. Field data collected by the U.S. Department of Energy’s (DOE) GATEWAY program for LED lamps
installed in museums shows that many have changed color beyond a reasonable tolerance well before
their rated lifetimes have been reached. Laboratory data collected by DOE’s CALiPER program between
2008 and 2010 reveals that many early LED products shifted beyond acceptable tolerances in as little as
a few thousand hours. In contrast, data from the L Prize® program illustrates that commercially available
LED products can have exemplary color stability that is unmatched by traditional light sources.
In addition to presenting data from the aforementioned DOE programs, this report discusses the metrics
used for communicating color shift, and provides guidance for end users on how to monitor
chromaticity and what to look for in manufacturer warranties. Also covered are the physical changes
that have been shown to lead to color shift in some types of LED packages. As with lamps and
luminaires, the data presented for LED packages shows that a wide variety of products are available; in
order for specifiers and consumers to make educated choices, more detailed and standardized
information is necessary.

Keywords
Light-emitting diode (LED), color shift, color stability, color maintenance, warranty
1 Introduction
The stability of a lighting product’s spectral output over time is a critical aspect of high quality lighting.
Changing spectral output—typically referred to as color shift—affects not only the color appearance of
the emitted light, but potentially also the color of objects the source is illuminating. Multiple light
sources that shift in variable ways can result in distracting patterns, and collective shifts can result in an
overall undesirable appearance. Either situation degrades the visual environment.
While in some applications large color tolerances may be acceptable, specifiers of lighting for inherently
visual environments, such as museums or patient examination rooms, must carefully consider the color
maintenance of light sources. The same can be said for applications that require numerous lamps or
luminaires to evenly illuminate a relatively monochromatic surface, such as a building façade or
architectural cove. If the output of a light source changes in color too much over time, it can no longer
be used, thus constituting a parametric failure of the product. Although just as important as lumen
depreciation (i.e., L70) or catastrophic failure, color shift is less frequently considered as a failure
mechanism.
The term color stability refers to a lamp’s ability to maintain a spectral power distribution over time. It is
separate from color consistency, which refers to the initial lamp-to-lamp variation in spectral power
distribution. Changes in spectral output resulting from changes to the ambient environment are also a
distinct issue, as is depreciation in radiant flux. However, all of these different issues may be
interrelated.
Color stability is a concern for all types of light sources, and performance varies between (and within)
different technologies. Halogen lamps are generally regarded as stable, whereas fluorescent lamps can
vary based on environmental conditions, but are generally stable in steady-state operation. Metal halide
lamps are notorious for poor color maintenance over time, although ceramic metal halide lamps may
offer some improvement. With increased adoption of LED products—and increasing hours of use—
awareness is growing that color shift may be an issue for some products. While improvements are being
made at the package and complete product levels, there is little publicly available information on the
extent to which LED products can be expected to maintain their spectral power distribution over time,
especially beyond 6,000 hours of operation.
This report provides color shift data for LED lamps used in real-world applications, as well as for LED
products monitored in a laboratory environment. It also includes some color shift data for example LED
packages. The document describes the known mechanisms of color shift, and provides end users with
guidance on how to monitor performance, as well as what to look for in a product warranty. Finally, it
looks at the development of new standards that may help manufacturers understand and communicate
expected performance over time.

1
2 Background
The color of a light source is often described using
chromaticity coordinates, a basic principle of the CIE
system of colorimetry [CIE 2004], or a metric derived
from them. The chromaticity coordinates of a source
provide a numerical representation of the color of the
light, but offer little indication of how the source will
render specific object colors—changes to the
chromaticity coordinates of a light source will lead to
changes in the appearance of rendered objects.
Chromaticity diagrams are relative plots of hue and
saturation—regardless of lightness, which is discarded
as a third dimension. In essence, they are two
dimensional representations of a three-dimensional
color space that is derived from color matching
functions. Subsequently developed three-dimensional
color spaces (CIELAB and CIELUV) are more appropriate
for describing object colors and absolute appearance,
whereas chromaticity diagrams are sufficient for
describing the output of light sources.
As the CIE system of colorimetry has developed, the way
in which chromaticity is plotted has been transformed
to create a closer match to visual evaluations of color
difference. That is, the plotted distance between two
sets of color pairs with equal visual difference is
intended to be the same. The three most commonly
referenced chromaticity diagrams are the CIE 1931 (x,
y), CIE 1960 UCS 1 (u, v), and CIE 1976 UCS (u', v'), as
shown in Figure 1. While (x, y) coordinates are most
frequently reported, color shift is most appropriately
documented as the difference or change in (u', v')
coordinates, written Δu'v', because the (u', v')
chromaticity diagram is the most visually uniform.
Nonetheless, no available chromaticity diagram is a
perfect representation of human color perception,
which itself varies from person to person. For many
years, research has been ongoing to improve the
methods used for documenting color difference, but as
with many lighting concepts, older methods remain
important for commerce.
Correlated color temperature (CCT) and Duv are Figure 1. CIE 2° chromaticity diagrams. The 1960 and
1976 versions are linear transformations of
measures of light source color appearance derived from the original 1931 version. The colored
backgrounds are shown for orientation only.
1
UCS is an abbreviation of Uniform Chromaticity Scale.

2
chromaticity coordinates. CCT and Duv values are calculated using the CIE 1960 (u, v) chromaticity
diagram (see Figure 2) due to the timing of the development of CCT, although the concepts are
applicable in any chromaticity diagram. Either measure alone cannot accurately describe the exact color
appearance of a source, but when used together the measures can be thought of as a two-dimensional
coordinate system with (somewhat) visually meaningful axes. In rough terms, CCT describes a yellow-
blue axis, and Duv describes a red-green axis. Nonetheless, reporting color shift as the change in one or
both of these values is insufficient for characterizing the magnitude of the perceptual difference,
especially since the (u, v) chromaticity diagram is not as perceptually uniform as its successor, (u', v').
Chromaticity “bins” demarcate areas on a chromaticity diagram, and are often used by manufacturers to
sort products and convey nominal performance. Quadrangular bins for nominal CCT are defined by the
American National Standards Institute (ANSI) Lighting Group and National Electrical Manufacturers
Association (NEMA) document C78.377-2011 [ANSI 2011], as shown in Figure 3. These quadrangles only
roughly represent color appearance, and the variation within a bin is noticeable and unsatisfactory for
almost all applications. That is, a source at one edge of the bin will appear very different from a source
at the other end. Accordingly, manufacturers often establish much finer bins for sorting their products.
Like with CCT and Duv—as well as any color rendering metric, such as the Color Rendering Index (CRI)—
color bins have little relevance to color shift, other than being gross representations of light sources with
different color temperatures.

Color Difference Concepts

Color difference and the limits of human color tolerance are complex topics that receive considerable
attention in research communities. Modern work in this field started with MacAdam ellipses, and
although alternative concepts are now available, the principles are often related the original ellipses.
The Δu'v' formula is currently recommended for describing color difference or color shift for light
sources, and many variants of ΔE (e.g., ΔE*ab, ΔE*94, ΔE*00, or ΔE*CMC) are used to describe color
difference of objects.
MacAdam Ellipses
In lighting literature, color difference is frequently described using MacAdam ellipses, although they are

Figure 2.
A close-up of the CIE 1960 UCS,
showing lines of constant CCT, which
were designed to be perpendicular to
the black body locus in that particular
diagram. At a given CCT, a source with
a positive value for Duv has a
chromaticity that falls above the black
body locus (appearing slightly
greenish), whereas a source with a
negative value for Duv has a
chromaticity that falls below the black
body locus (appearing slightly pinkish).
The lines in this chart represent a Duv
range of ± 0.02, which is much greater
than ANSI tolerances for white light.

3
widely misunderstood. MacAdam ellipses were developed under strict experimental procedures, with
the ellipse representing a standard deviation of color matching (SDCM) for a single highly trained
observer who made repeated matches [MacAdam 1942]. These ellipses were derived throughout the
1931 (x, y) chromaticity diagram, and the conspicuous difference in size influenced subsequent
chromaticity diagram revisions (see Figure 3). As shown in Figure 4, in the (u', v') chromaticity diagram,
the ellipses are approximately circles, with a one-step ellipse having a radius of about 0.001—at least in
the region around the black body locus (nominally white light).
MacAdam ellipses are often reported with a step size (e.g., seven-step), which is in reference to the
multiple of the original SDCM ellipse. In this way, two pairs of chromaticity coordinates can be said to
represent an x-step difference. Importantly, MacAdam’s ellipses characterized the difference between
the point at the center of the ellipse and a point at the edge of the ellipse, not two points at the edge.
Thus, two points at opposite ends of a seven-step ellipse are correctly characterized as exhibiting a 14-
step difference (see Figure 3).
A common misinterpretation of MacAdam ellipses is that they are universally related to a just-noticeable
difference. Under the conditions of MacAdam’s experiment, a three-step difference was determined to
be just noticeable [Wyszecki 1982], but that inference cannot be extrapolated to other environmental
conditions that are drastically different from the small field of view under which the research was
conducted. The noticeability of a difference in chromaticity depends on many factors; for example, two
adjacent lamps aimed at a white wall might appear to be different colors, but when used in downlights
at the opposite ends of a room, the difference might not be apparent. Further, as is the case with all
lighting metrics, chromaticity and color difference formulas are based on average observers, so
individual perceptions may vary.

Figure 3. Left: 10-step MacAdam ellipses plotted in the 1931 (x, y) chromaticity diagram. The different sizes of the ellipses
illustrate the nonuniformity of the diagram.
Right: Enlarged image of a typical ellipse. An x-step difference is always relative to the target chromaticity
(center of the ellipse). Point A is one step different from the target, but two steps different from point B. In other
words, all points encompassed by an ellipse (e.g., points C and D) are not within the specified step size tolerance
of each other, but only with n the step size tolerance of the center point.

4
Figure 4. ANSI tolerances for CFL and SSL sources in the 1976 (u', v') chromaticity diagram. For this diagram, MacAdam
ellipses are approximately circular around the black body locus, with one step equal to approximately 0.001.

Δu'v'
Determining MacAdam ellipses requires several formulas, whereas Δu'v' is simply calculated as the
Euclidian distance between a pair of chromaticity coordinates in the (u', v') chromaticity diagram. The
relationship to MacAdam ellipses in the region around the black body locus—as previously described—
allows for the approximation that a Δu'v' value of 0.001 is equivalent to a one-step difference. As with
MacAdam ellipses, however, the noticeability of a difference is dependent on many factors, so there is
no universal limit for what is considered an acceptable Δu'v'.
Given the mathematical simplicity of specifying color shift limits in terms of Δu'v' and the rather complex
procedure for calculating a MacAdam ellipse—not to mention the many common misconceptions—both
manufacturers and specifiers may benefit from using the Δu'v' terminology. One important fact to
remember, however, is that neither MacAdam ellipses nor Δu'v' communicate anything about the
direction of the color shift, which may or may not be important. Rather, Δu'v' quantifies the total color
difference that may be the result of changes to either Duv or CCT (or any other two-dimensional
representation of chromaticity), as shown in Figure 5.

Existing Standards for Measuring and Documenting Color Shift

The Illuminating Engineering Society’s (IES) LM-80-08, Measuring Lumen Maintenance of LED Light
Sources [IES 2008], prescribes an approved method for long-term measurement of LED packages—but
not complete LED lamps or luminaires (that document is in progress). While it primarily targets lumen
maintenance, LM-80-08 requires that chromaticity shift be measured (and subsequently reported) over
the course of the test procedure, which must last a minimum of 6,000 hours. There are no specified
acceptable limits for color shift, and there is no method for projecting future color shift as there is for

5
 Figure 5. Close-up of the 1976 (u', v') chromaticity
diagram illustrating the difference between
Δu'v', ΔDuv, and ΔCCT. ΔDuv and ΔCCT are
insufficient for characterizing color
difference. Only Δu'v' describes the total
color difference that might be observed,
although it does not describe the direction of
the shift. As such, Δu'v' (1) and Δu'v' (2)
would have the same value, despite shifting
in different directions.

lumen depreciation (IES TM-21-11 [IES 2011]). One important note with LM-80-08 is that the ambient air
temperature and solder point temperature of the LED package are at equilibrium, which is unlikely to
mimic real-world operating conditions.
The ENERGY STAR® program does mandate that Δu'v' at 6,000 hours of operation not exceed 0.007 [EPA
2012, 2013]. This is perhaps the only industry-wide criterion for color shift. It is a reasonable starting
point, but may not be strict enough to ensure very high-quality lighting, especially since the lifetimes of
LED products routinely far exceed 6,000 hours.
Although more critical for color consistency than color stability, it is important to note that the
quadrangles defined by ANSI/NEMA offer very large tolerances for nominal CCTs. For example, the
3500 K bin has corner points with a Δu'v' of 0.0266, or nearly four times the ENERGY STAR tolerance for
color shift. Needless to say, two lamps with the same nominal CCT designation can appear vastly
different, and the difference can be exacerbated by color shift over time. Many LED chip manufacturers
can offer much finer bins than the nominal CCT tolerances established in C78.377-2011.

6
3 Color Shift Data
The data presented here, which is conglomerated from several DOE efforts, collectively provides an
understanding of the current state of the market and helps to identify a path toward improving the color
stability of future products. However, the data has largely been analyzed post-hoc, and much of the
work was never intended to be a rigorous scientific investigation of color shift. It is possible that
analyzing different products under different conditions would provide different results. However, the
information presented illustrates both the high potential for LED lighting and some of the hurdles that
must be overcome.
DOE GATEWAY Program Data
The GATEWAY program has featured three museum projects, including a recent collaboration with the
Smithsonian American Art Museum in Washington, DC [DOE 2012]. Museums are particularly
demanding for lighting, especially with regard to color appearance and color maintenance. While
GATEWAY demonstration projects typically focus on the initial performance and expected payback of
LED lighting systems, the LED lamps installed at the Smithsonian were reevaluated after several
thousand hours of use when color shift became apparent.
Many different lamps were installed as part of the Smithsonian GATEWAY project—and in subsequent
additional trials. As some of the lamps have aged, they have shifted in visually noticeable ways, in some
cases requiring replacement well before the lamps’ rated lifetimes. While there was a strong tendency
for lamps that used the same LED chips and that were placed in the same environment to shift as a
group, it is common for manufacturers to use different chips depending on the specific lamp
performance (e.g., different luminous intensity distribution). Very small die used for narrow spot lamps
may shift differently from larger-die flood lamps from the same manufacturer—as was also experienced
at the Smithsonian. Importantly, the premature failures due to color shift happened independent of
lumen depreciation or catastrophic failure. Warranty coverage for color stability problems ranged from
manufacturers that explicitly provided coverage, to manufacturers that were willing to remedy the
problem once they received a complaint, to manufacturers that were unresponsive.
Because the initial intent of the Smithsonian GATEWAY project was not to document performance over
time, no procedure was followed to track the performance of any given lamp (i.e., to record longitudinal
data). Rather, individual lamps with varying hours of operation were analyzed post-hoc once color shift
issues were identified. That is, values for Δu'v' were calculated for aged lamps relative to new lamps of
the same type, rather than for the aged lamp versus itself when it was new. The lamps ranged up to
6,600 hours of operation, typically in a cylinder-style trackhead that sometimes had additional optical
elements such as a lens, louver, or screen in front of the lamp.
During installation, the Smithsonian evaluated all of the lamps to visually confirm that none was
noticeably different. Additionally, during specification, lamps were tested to confirm that the color was
uniform across the beam (color over angle). These actions add some credibility to the method of
comparing aged lamps to new lamps of the same type, although the data must still be used cautiously.
The GATEWAY program has also collected and measured lamps from the Hallie Ford Museum of Art at
Willamette University, an installation that was not part of a GATEWAY demonstration. The Smithsonian
and Hallie Ford projects are indicative of wider trends that have been reported throughout the museum
community, which serves as a vital test bed for LED technology.

7
Table 1. GATEWAY data on color from lamps used in museums.
Hours Relative
Type Description of Use Output CCT Duv Δu'v' Time2 Δu'v' Optics3
A1 PAR30, 10°, 2700 K 0 - 3063 -0.0032 - 0.0007
A1 PAR30, 10°, 2700 K 100 92% 3066 -0.0028 0.00051 No Data
A1 PAR30, 10°, 2700 K 5,750 100% 2949 0.0022 0.01001 No Data
A1 PAR30, 10°, 2700 K 6,600 97% 2990 0.0007 0.00701 No Data
A1 PAR30, 10°, 2700 K 6,600 90% 3020 0.0005 0.00601 No Data
A2 PAR30, 25°, 2700 K 100 - 3008 0.0005 - 0.0003
A2 PAR30, 25°, 2700 K 6,600 94% 2922 0.0016 0.00521 0.0003
A2 PAR30, 25°, 2700 K 6,600 95% 2928 0.0023 0.00581 No Data
B1 PAR38, 10°, 2700 K 0 - 2727 -0.0020 - 0.0009
B1 PAR38, 10°, 2700 K 4,000 103% 2981 0.0008 0.01101 0.0009
B2 PAR38, 10°, 2700 K 0 - 2744 -0.0024 - 0.0008
B2 PAR38, 10°, 2700 K 4,000 91% 2687 0.0029 0.00881 0.0005
C MR16, 2700 K 0 - 2789 -0.0024 - 0.0013
C MR16, 2700 K 5,000 78% 2548 0.0054 0.0183 0.0004
D MR16, 2700 K 480 - 2771 -0.0008 - 0.0017
D MR16, 2700 K 2,400 83% 2655 0.0004 0.0062 0.0022
D MR16, 2700 K 2,400 81% 2754 0.0043 0.0076 0.0021
1. LM-80 data suggests a maximum shift of 0.035 at 85°C and 1000 mA drive current. The package used for product types C
and D are unknown.
2. Δu'v' was calculated relative to the baseline case, which was an unused version of the same lamp type (i.e., the first sample
in each group). Red values exceed the ENERGY STAR criterion of 0.007.
3. For each individual lamp with secondary optics removed compared to the same lamp before the optics were removed.

Table 1 provides data for a sample of lamps that the GATEWAY program has evaluated. These lamps
were measured according to IES LM-79-08 at the Pacific Northwest National Laboratory (PNNL)
photometry lab. They were operated and measured in open air (i.e., bare lamp). Each entry is for a
unique product, with the approximate hours of use listed. Color shift was calculated relative to the first
listed value for each product type. Many of the lamp samples had a Δu'v' greater than 0.007 relative to
the baseline sample, all in 6,600 hours or less.
Although the data was not collected from the same lamp samples over time, it is highly suggestive of
substantial color shift, especially because the initial chromaticity coordinates were supported by
manufacturer-supplied data, if available. In the subsequent analysis, the data is treated as
approximating longitudinal data. Several plots of 1976 (u', v') chromaticity diagrams are shown in Figure
6. These charts show that the measured shifts are primarily—but not always—toward a more yellow
appearance.
In addition to evaluating bare lamps, secondary optical systems (e.g., plastic lens elements) were
removed in order to isolate the performance of the LED package. Although at least one of the products
showed visible discoloration of a physical material over time, removing the optical systems resulted in
minimal changes to the measured chromaticity. 2 Further, there was little difference in the total shift

2
Upon further examination, the discoloration was of the white board on which the LEDs were mounted, rather than the lens.

8
when removing the lens from the newer lamps or older lamps, as shown in Figure 6. Thus, it is
reasonable to conclude that the shifts are a result of changes to the LED packages.

Figure 6. Color shift of lamps used in GATEWAY demonstration projects, as shown in the 1976 (u', v') chromaticity
diagram. Each point is labeled with its hours of use. The data is not longitudinal.

9
Importantly, the color shift observed in these lamps was not predicted by LM-80-08 test data (if
available), which would suggest shifts at 6,000 hours of no more than Δu'v' = 0.0035, depending on the
exact package and operating conditions. At least in the case of lamp manufacturer A, the lamp and
luminaire combination was approved (by the lamp manufacturer) after thermal testing in the
manufacturer’s laboratory before operation commenced, and field measurements of ambient
temperature did not seem to be outside the range for LM-80-08 testing. This lamp also employed a
thermistor to prevent thermal conditions from exceeding an acceptable range, thus preventing damage
to the lamp (note that all data shown here was collected according to LM-79-08, so the output should
not have been changed by the thermistor at the time of measurement). Additionally, the color shift was
similarly unacceptable whether a glass lens was used or just an open louver, although this may have
caused little difference in thermal environment because the trackheads had an enclosed back.
Figure 7 shows the change in SPD for lamp types B1 and B2. The 4,000-hour sample for lamp type B1
was generally more blue than the baseline sample, and emitted about 3% more lumens—again, the data
is not longitudinal, meaning two different lamps with different hours of use were measured, and they
did not necessarily start out the same. The blue peak in the aged SPD of lamp type B1 had more energy,
whereas the yellow peak had less energy and was shifted to a slightly shorter wavelength. In contrast,
lamp type B2 shifted toward yellow and emitted about 9% fewer lumens. Accordingly, the decrease in
the amplitude of the blue peak was greater than the decrease in amplitude of the yellow peak.
Lamp types A1 and A2 were also measured in a sample trackhead of the same type used when they
were operated in the field, both with and without an auxiliary glass lens (data not shown on plots). In
both cases, the minimal color difference between the test condition and baseline was much less than
observed based on hours of use, with about half attributable to operation in the luminaire and the other
half attributable to adding the lens to the luminaire. In general, this shift was comparable to that
measured when the secondary optical system was removed from the lamp, and was likely caused by
changing the ambient temperature of the air immediately around the lamp.
DOE CALiPER Program Data
Between 2008 and 2010, the CALiPER program conducted long-term testing of approximately 50
complete LED lamps and luminaires at an independent photometric lab. While many improvements
have been made to LED products in the past three years, this data is nonetheless valuable for
understanding some of the mechanisms of color shift that remain a potential problem. A subset of the
original data, which removes some products considered especially outdated (e.g., those using 5 mm
LEDs) or that were tested using a different protocol, is included here.
After undergoing initial LM-79-08 testing in an integrating sphere, the products were mounted in an
apparatus to allow for continuous operation at an ambient temperature of 25°C (±5°C) and constant
input voltage. Spot illuminance and chromaticity measurements were made every 500 hours using a
handheld meter with a jig and masking apparatus for consistency. After 6,000 hours, the products were
again measured in the same integrating sphere. Some products failed prior to 6,000 hours, and others
underwent an additional 6,000-hour cycle. 3
Figure 8 shows the initial and final chromaticity coordinates of the products, as measured in an
integrating sphere according to LM-79-08. The products plotted in red underwent an additional 6,000
hours of testing, during which the color shift generally continued in the same direction. As shown, many
of the products shifted along the blue-yellow axis, although there were a few exceptions. Products with

3
Four lamps had an initial cycle of 7,500 hours instead of 6,000 hours.

10
Figure 7. Spectral power distributions for new lamps compared to lamps that had been operated for 4,000 hours. Lamp
type B1 was more blue at 4,000 hours than a new lamp of the same type, whereas lamp type B2 was more yellow.

a higher CCT were much more likely to exhibit greater color shift. In total, 15 of the 45 products listed
had a Δu'v' of more than 0.007 in the first 6,000 hours of operation. Other interesting observations
include:

• Of the 15 products with a Δu'v' exceeding 0.007, eight had a CCT above 4500 K. Only two of the
ten products with a CCT above 4500 K did not exceed the ENERGY STAR Δu'v' limit of 0.007 at
6,000 hours of operation (the lamps were not necessarily ENERGY STAR qualified, but that
threshold is used as a reference).

11
Figure 8. Initial and final chromaticity of lamps tested by CALiPER in 2008–2010.

• Of the seven products using an RGB (red, green, blue) or hybrid approach to making white light,
two exceeded a Δu'v' of 0.007, a similar proportion to the total group.

• Of the seven products that underwent a second 6,000-hour test period, six had a Δu'v'
exceeding 0.007 by the end; only two exceeded that threshold after the first period. These
products were the oldest—and some of the earliest—LED products on the market, purchased by
CALiPER in 2006 or 2007.

• The most extreme shift was for an undercabinet luminaire, which had a Δu'v' of 0.045 at 7,500
hours. Several other products exceeded a Δu'v' of 0.020, which is nearly three times the ENERGY
STAR limit and likely unacceptable for any lighting application.

12
Figure 9 shows data for a subset of the products shown in Figure 8 that had intermediate measurements
taken every 500 hours using a handheld meter. As shown, a product’s chromaticity may not follow a
continuous, straight path from the initial measurement to the 6,000-hour measurement. The behavior
of the products was often erratic, sometimes reversing course in the middle of testing. This pattern
suggests that more than one mechanism of color shift may be at work. However, it is important to note
that this data was collected using a less precise meter and in an environment with less rigorous control.
The environment may more closely resemble typical operating conditions, but it can make it difficult to
separate out color shift caused by thermal changes from color shift caused by physical changes over
time.

Figure 9. Chromaticity over time for a subset of the lamps shown in Figure 8. When intermediate points are shown, the
color shift appears much less predictable.

13
DOE L Prize Program Data
The L Prize-winning A lamp has become one of the most-tested LED products to date. As of April 29,
2013, a large sample of prototype lamps (n = 202) had undergone 25,000 consecutive hours of operation
at a target maintained ambient temperature of 45°C. One piece of the data collected at regular intervals
during that time was the spectral power distribution—and thus chromaticity and color shift.
Notably, the L Prize lamp uses remote phosphor technology, in which the phosphor used to create white
light is not incorporated directly into the LED package. A minority of LED products uses this approach,
and many of the color shift mechanisms discussed in the analysis section of this report are only
tangentially relevant to remote phosphor lamps. Additionally, the samples tested were specially built for
the evaluation rather than commercial sale. Nonetheless, these data are included to highlight the
potential of LED technology.
Contrary to the GATEWAY products and some of the CALiPER products included in this report, the L
Prize-winning lamp has shown remarkably little color shift over its lifetime. Although always minimal,
the precise magnitude of the shift depends on the analysis method. For this report, the baseline value
for each sample was the measurement taken at 22 hours of operation. Additionally, the average Δu'v'
value was calculated as the mean of Δu'v' for each lamp, rather than the Δu'v' for the mean chromaticity
coordinates at each measurement interval. Selecting a different baseline measurement or using a
different averaging method would not alter the general conclusions, even if the precise values differ
somewhat.
One important consideration is the dependence of chromaticity on ambient air temperature. Although
the L Prize lamps were operated in a special environmental chamber, some variation did occur over the
25,000-hour measurement period. Specifically, there were a few measurement points that clearly
appeared to be anomalies, as they were substantially different from the preceding and proceeding
measurements. For example, the average Δu'v' for measurements taken at 17,106 hours—when the
temperature of the sphere dipped approximately 4°C—was approximately 0.001 higher than the mean
measurements immediately before and after. The temperature dependence of the color shift
measurements is shown in Figure 10. This attribute is important to understand when attempting to

Figure 10. Mean Δu'v' over time versus the temperature of the integrating sphere used for measurement. The
temperature at the sphere does not identically match the temperature of the air surrounding the lamps, due to
stratification within the apparatus, but the relationship is still apparent.

14
make field measurements of color shift, and also provides a baseline for performance expectations in a
real installation. Measurements deemed to be influenced by thermal anomalies were removed from
subsequent analyses; in charts, those areas are shaded gray.
Excluding times of thermal anomalies, average Δu'v' values remained at or below 0.001 for the entire
duration of measurement, as shown in Figure 11 (black line). The highest recorded Δu'v' for any of the
lamps was 0.0047 (orange line), although this was a large spike that was not seen for any of the other
samples. The Δu'v' for that sample, Q7, was less than 0.001 at 25,000 hours. The lamp sample with the
highest mean Δu'v' and highest value at 25,000 hours was F4 (red line). This performance was also
atypical, with only one other lamp (L5) performing similarly. In contrast, lamp C12 (blue line) had the
lowest mean Δu'v', and a maximum Δu'v' of just 0.0006, which is more than 10 times lower than the
ENERGY STAR criterion. Further, sample A12 (green line) had a Δu'v' at 25,000 hours of just 0.0002, the
lowest of any of the samples—although it did drift slightly away from its start point in the first third of
the measurement period before coming back.
Although it is not the subject of this report, it is important to understand the color consistency of the L
Prize lamps, as it gives context to the color shift data. Figure 12 shows the chromaticity coordinates of
all 202 lamp samples at the baseline and 25,000-hour measurements. Although not numerically
quantified, it is clear that the trend is for shifts slightly toward blue. Also important is the fact that the
magnitude of difference within either the initial or final chromaticity measurements far exceeds the
change in chromaticity measured for any of the lamps. That is, the spread of initial or final chromaticity
coordinates, with a Δu'v' for the extreme points of approximately 0.006, is greater than for any of the
shifts measured over time for an individual lamp.
Figure 13 shows the chromaticity of six individual L Prize test samples that have different shift
characteristics. Although the overall trend was a shift toward blue (lower u' values), this was not the

Figure 11. Δu'v' for the best and worst performing lamps. Gray areas indicate data that has been removed due to thermal
anomalies.

15
Figure 12. Baseline and final (25,000-hour) chromaticity coordinates for each lamp sample, shown at two different scales
of the same plot.

16
Figure 13. Chromaticity coordinates for six of the sample L Prize lamps. The baseline measurements are shown in blue, and
subsequent measurements in red.

*The plot for sample F4 does not have the same axes.

case for all samples—the starting point is marked by the larger blue circle. The exact direction and
magnitude of the shift was variable, although it was predominantly along the u' axis (approximately
blue-yellow).

17
Importantly, a tolerance must be applied to the measurement device, as well as to the thermal stability
of the apparatus. This measurement noise is difficult to quantify, but it likely increases the observed
level of variation. Undoubtedly, the color stability of the L Prize lamp is very good and illustrates the high
potential for LED products.

18
4 Discussion
Mechanisms of Color Shift: Operating Conditions
LED product characteristics—including luminous flux, chromaticity, and efficacy—are dependent on
operating conditions. LED packages are typically tested and binned by the manufacturer at a specific
temperature (e.g., 25°C or 85°C) and specific drive current (e.g., 350 mA). If the LEDs are used in an
application where the temperature and/or drive current differs from what was used during the product
testing, then the chromaticity can be different. The magnitude of chromaticity shift with temperature
and current changes will differ from one manufacturer to another, as the technologies used to fabricate
the LEDs will each be different. In some cases, the shifts are so small as to be unnoticeable, whereas in
other cases the shifts are very significant.
For LED products, color shifts caused by changes in the operating conditions are generally recoverable.
That is, if the change is reversed, the chromaticity should return to its original point. This is not unlike
some conventional technologies, such as fluorescent, and there is probably little opportunity to alter the
behavior. While users should be aware of color shift due to changes in operating conditions, the greater
concern for the lighting industry is reducing or eliminating changes that stem from premature physical
degradation. The data presented in this paper were all for lamps in steady-state operation, and the
ambient conditions did not substantially change.
It should be noted that there might be long-term consequences from operating at a higher temperature,
because heat generally increases the degradation of materials. Similarly, on-off cycling—and the
associated thermal changes—may have some unrecoverable effect on the physical materials of an LED
package. Such an effect would not be accounted for in LM-80-08 testing of LED chips. In short, changes
due to operating conditions and physical changes causing color shift are not entirely independent.
Another aspect of color shift is its relation to dimming. LED products may or may not change
chromaticity when dimmed, and this behavior may or may not be intentional or desired. In general, this
is a separate issue from color stability at steady-state operating conditions.
Mechanisms of Color Shift: Physical Changes to LED Packages
The most commonly available LED packages are a combination of a “blue” LED and “yellow” phosphor—
in reality, the blue LED emits energy over a narrow range of wavelengths, while the phosphor
downconverts some of the emitted blue to radiant energy having a wide range of wavelengths, including
those perceived as green, yellow, orange, and red. The combination appears nominally white. These LED
packages are often referred to as phosphor-coated (PC) LEDs. Other LED packages or arrays use a
combination of discrete LEDs (e.g., red, green, and blue) that are mixed to create white light, and still
others use a hybrid approach that adds an additional discrete LED (usually red) to a PC-white LED
package.
The focus of the subsequent discussion is on typical PC LEDs. Many of the issues discussed here also
apply to hybrid or color-mixed LED packages, although the additional diode(s) may add another
mechanism(s) of instability. Specifically, the different colored diodes can each degrade differently over
time. Integrating the output from multiple colored LEDs can also pose a challenge for color uniformity,
especially for products with narrower distributions. Remote phosphor products, like the L Prize lamp,
are fewer in number and likely have different mechanisms of color shift. The L Prize lamp is one carefully
engineered example of a remote phosphor lamp with excellent color stability, but as with other product
types, variation in performance is likely substantial.

19
PC LEDs come in a wide variety of physical forms that have consistently evolved as LED technology has
matured. Long-term performance can be difficult to assess, given the length of time needed for
measurements, and significant problems may only arise after prolonged field use. At the package level,
several changes can occur that result in the emitted light typically shifting toward either yellow or blue.
The exact mechanism is primarily based on the type of package. The following are several examples, but
the list is not exhaustive and there is large variability from manufacturer to manufacturer and product to
product:
 Some (early-generation) LED packages mixed the phosphor with a soft silicone and dispensed it
into a reflector cavity (Figure 14). Over time, the phosphor particles were found to settle,
allowing more blue photons to be emitted from the package without conversion to longer
wavelengths. The result was a severe shift toward the blue end of the spectrum.
 Most low- and mid-power plastic LED packages use a similar mixing method to the one
described previously, but the silicones used in these products tend to be much firmer. As a
result, the particles do not settle in the package; however, there is potential for color shift to
result from discoloration of the plastic (Figure 15). Polyphthalamide (PPA) is a common plastic
resin used in LEDs that is known to discolor at high temperatures and when exposed to short-
wavelength radiant energy.
 Another type of LED package coats the phosphor in a thin layer on the top surface of the chip,
with silicone molded over top (Figure 16). With one type of coating technology, the phosphor
layer can curl up at the edges of the chip, resulting in blue photons being emitted at high angles
without conversion to longer wavelengths (Figure 16a). A shift toward blue is then observed.
With other coated-chip products, delamination of the phosphor from the surface of the chip can
occur at high temperatures (Figure 16b). The resulting air gap between the chip and the
phosphor increases the angle at which the blue photons pass through the phosphor, and the
resulting mean free path of the blue photons through the phosphor increases, causing a shift
toward yellow. Better attachment methods in more recent products have helped to alleviate
these problems. Color shift characteristics for several coated LED packages, both older and
newer, are shown in Figure 17.
 In some of the newest LED packages, phosphor is applied to the entire package surface,
including the chip and substrate (Figure 18). This can further alleviate the curling and
delamination problems, greatly increasing color stability. As with any of the package types
described in this report, there is substantial variation from one product type to the next; the
example data provided does not represent every product. Further, the data does not represent
an endorsement of any package technology, but is simply meant to illustrate the wide variety of
products that are available and the potential for color stability issues.

Figure 14. For some early-generation LEDs, phosphor mixed with silicone settled over time.

20
Figure 15. Some low- and mid-power LED packages can shift color because the plastic housing discolors. The diagrams
above illustrate that some of the light emitted from the die that would reflect off the housing may be absorbed or
shifted if the plastic discolors, thus changing the overall emission from the package. The plot below shows the
average shift for a sample of one type of product. The ENERGY STAR criterion is shown for reference only.

LED package technology is constantly changing and improving, with the ultimate goal of a package that
is physically stable from start to end. However, the latest and greatest LEDs do not make it into every
final product, and there is often a substantial lag between chip development and deployment in end-use
lamps and luminaires. It can be very difficult for end-users to know what generation or type of LED is
included in a product without asking manufacturers detailed questions. This issue is exacerbated when
manufacturers make subtle changes to improve performance but do not re-label the new products to
indicate a new design.
Adding another layer of complexity beyond changes to the packages, the materials used for primary or
secondary optics may change over time. In some products (e.g., 5 mm radial lamps), epoxy is used as an
encapsulant. Over time, discoloration of the epoxy is frequently observed. Likewise, manufacturers must
carefully select the materials used for optical components or risk discoloration, because many
chemicals, especially adhesives, used in the manufacturing of lamps and luminaires can be unstable
during operation. In many cases, the lamp or luminaire manufacturer is different from the LED package
manufacturer. This can make understanding the cause of observed color shift more difficult, and
warranty claims more complicated.
Physical changes may also occur in the driver, which could result in a different drive current, which in
turn might cause color shift. This issue has not yet been extensively investigated; however, it may be a
contributing factor to differences between LM-80-08 color shift data for LED chips and the color shift
data for complete products described in this report.

21
 Comparisons to Conventional
Products
Color consistency and color stability
can vary greatly with different
technologies. Incandescent lamps,
including halogen variants,
generally have good color stability,
and some specifiers and users like
the fact that they shift to a warmer
color when dimmed. In contrast,
metal halide lamps are notorious
Figure 16. Curling (A) and delaminating (B, close up) are two potential for having very unstable color.
causes of color shift with phosphor-coated LED packages. Besides substantial shifts early in
Curling generally causes a shift toward blue, whereas life, chromaticity may change on
delaminating generally causes a shift toward yellow.
the order of Δu'v' = 0.020 over a
lamp’s lifetime [NLPIP 2004].
Ceramic metal halide lamps offer
some improvement. Due to their
poor color stability, metal halide
lamps are rarely used in applications
where color quality and color
stability are critical.
In general, color stability for
fluorescent lamps is considered very
good, with typical changes over a
majority of life less than a Δu'v' of
0.004. However, fluorescent lamps
do not always have good color
consistency; that is, fluorescent
lamps from different manufacturers
with the same listed CCT can have a
wide range of measured
chromaticities. More importantly,
output is strongly affected by
ambient temperature.
Fluorescent lamps are almost
exclusively used for general ambient
lighting, where color tolerances are
not always high and where slight
variations among lamps are less
noticeable because the light from
many lamps is mixed together in the
Figure 17. Top: Some older phosphor-coated LED packages exhibited
final lighted environment. Many of
rapid color shift after several thousand hours of operation. the tolerances established for LED
Bottom: An example of a newer phosphor-coated LED package lamps came from previous
with very good color stability. Both plots show the average tolerances established for
color shift for a sample of lamps.

22
Figure 18. Average color shift for a newer type of LED package that has been measured to have very good color stability
under several different operating conditions.

fluorescent lamps; for example, the IES/ANSI quadrangles defined in C78.377-20111 were derived from
seven-step MacAdam ellipses used to define nominal CCTs for compact fluorescent lamps (ANSI
C78.376-2011) [ANSI 2001].
There has been some debate over the need for more stringent color tolerances for LED products than
for incumbent conventional technologies —from ENERGY STAR or other programs. This is driven largely
by the lofty potential of LED products, rather than by actual user experiences to date. In general, stricter
tolerances for color stability may only be needed in specific applications, and may not be needed for
most general lighting.
Warranty Issues
Although color shift is a known cause of parametric failure for LED sources—much like lumen
depreciation—it is covered by only a small number of manufacturers’ standard warranties. This could
partially be attributable to the difficulty in predicting color shift from measured data, and as a result,
many existing warranties cover only the period of LM-80-08 measurement (e.g., 6,000 hours). As shown
in Figures 15 and 17, however, color shift may only become a problem after that time. Nonetheless, at a
minimum, manufacturers should warranty that Δu'v' should not exceed 0.007 at 6,000 hours, especially
for ENERGY STAR-qualified products. This is likely the minimum quality standard for any interior
architectural lighting.
One consideration with color shift warranties is what to use as the baseline measurement from which
Δu'v' can be calculated. Moreover, it must be established whether the color shift is relative to a baseline
for each lamp or is relative to other lamps of the same type operated for the same number of hours. In
other words, is the situation where all lamps change in the same way considered a failure? Finally,
manufacturers should clearly state any restrictions that apply, such as operating a lamp in an enclosed
luminaire, or above a certain ambient temperature. If restrictions are based on the junction
temperature not exceeding a certain limit, the manufacturer should communicate a procedure by which
users can ensure they are meeting the terms of the warranty. For example, some manufacturers provide
a temperature measurement point (Figure 19) on the lamp and indicate the maximum allowable

23
 temperature at that point.
This information is helpful to
fixture manufacturers that
must design appropriate
airflow or other forms of
thermal management.
Best Practices for
Monitoring Color Shift
Often, potential color shift is
overlooked at the time of
installation, with concern
arising later, when there is
visible evidence of change.
However, it is very important
to begin monitoring color at
the time of installation, or at
least take actions to allow for
investigation at a later time.
The simplest, and crudest,
way for consumers to track
color change is to keep a
minimum of two products in
storage as a reference for
post-hoc measurements. It is
important to make sure that
all products, including the
reference(s), match color as
Figure 19. A temperature measurement point, marked Tc, is shown on an example closely as is required for the
lamp.
application at the time of
installation. It is also
important to record the installation date for each lamp, track or estimate hours of use, and keep the
invoice showing where and when the products were purchased.
In critical applications, like museums, or for scientific demonstration projects, a proactive approach may
be more valuable, especially if warranties include color shift as a failure mechanism. Ideally, regular
measurements would be taken of the same product according to LM-79-08; however, this would likely
be prohibitively laborious and expensive. Alternatively, an initial LM-79 measurement could be used to
establish a baseline. 4 Subsequently, spot measurements with a handheld meter may be used to track
changes. It is likely that the measurements made with the handheld meter will not exactly match those
made using an integrating sphere and following the LM-79-08 procedure; 5 however, these spot
measurements may help to identify if and when an unacceptable color shift has occurred. At this point,
a follow-up measurement can be made in a photometric laboratory, if necessary.

4
Because the output can change substantially in early life, taking a baseline measurement at 0, 100, and 1,000 hours may be a
preferred approach, if possible.
5
Both differences in operating conditions (e.g., ambient air temperature, lamp tilt) and differences in metering equipment may
contribute to differences in measured chromaticity.

24
In many cases, even one LM-79-08 measurement may not be feasible—or beyond the budget of the
specifier or building owner. In such cases, monitoring chromaticity with a handheld meter or comparing
used lamps with new lamps may be the only options. These practices can help identify a problem, but
their relevance to making a warranty claim should be addressed with the manufacturer prior to
purchase.
Future Developments
As with many aspects of LED performance, there is substantial need for accelerated testing to predict
color shift. The reliability of such calculations is presently a concern, although the IES has initiated a
committee that will be charged with developing an approved procedure. This work is targeted toward
LED packages, for which the exact operating characteristics are more easily controlled. Much like
lifetime and reliability, extending component-level data to complete LED lamps and luminaires can be
difficult.
In the meantime, the lighting industry should pay increased attention to the issue of color shift, which
could lead to unacceptable levels of end user dissatisfaction with the technology. It is clear that the
color stability of LED packages is improving, but more work is needed to communicate performance
between different segments of the industry. Greater education on the metrics and language used to
describe color shift would help to facilitate higher-level discussions.
As the issue of color shift has become more widely known, some manufactures are beginning to respond
by expanding warranty coverage. However, at this time there is no standardized threshold for
acceptability—such as L70 for lumen maintenance—beyond what is specified by qualification programs
like ENERGY STAR. Furthermore, there is no standardized method for users to measure or document
color shift, which may make the warranty claim process confusing or overly burdensome. In the future,
harmonized warranty procedures and coverage may help alleviate specifier concerns stemming from the
large variation between different products, and provide assurance until better-performing products
permeate the market.

25
5 Conclusions
Improving the long-term color stability of LED products is highly desirable, lest they be relegated to
limited applications where color shift is not a concern. Data collected by multiple DOE SSL programs
suggests that the performance of some products is closer to that of metal halide lamps than to
fluorescent or halogen lamps, but also suggests that some LED products offer excellent color stability.
Many factors contribute to the variation in color maintenance performance of LED lamps and
luminaires, including the construction of the LED packages themselves. In the past few years, several
generations of packages have been developed by numerous manufacturers. Over that time, much has
been learned about the long-term color stability of products, and improvements have been made.
However, the newest generations of LED packages do not have full market penetration, nor are they all
immune to color shift. Thus, end-users who are concerned about color stability must be especially
careful in choosing a product; however, it is difficult to predict which products might have better color
stability, even when one is aware of the issue and well versed in the technical details.
As GATEWAY and CALiPER data have shown, color shift is especially complicated at the complete
product level, with numerous contributing factors. As a result, in applications where color quality is
important, some products have useful lifetimes that are substantially shorter than their rated lifetime,
which may in turn affect the anticipated payback. These concerns could be alleviated with manufacturer
warranties, but there are few currently available warranties that cover color shift.
Future work on color shift should include:
1. Continued development of LED package technology to improve color stability.
2. Development of standards for predicting long-term color shift performance from a more limited
set of measured data.
3. Investigation of the interaction of multiple factors affecting the color stability of complete LED
lamps and luminaires.
4. Education within all segments of the lighting industry regarding the issue of color shift and the
tools to communicate it properly.
5. More widespread inclusion of color shift in product warranties, pursuant to an agreed-upon
method for documenting color shift beyond an established threshold.
As more and more LED products are installed in buildings and the hours of use continue to grow, long-
term performance issues such as color shift will become more prominent. While it may be convenient to
overlook color stability in favor of improved efficacy or lumen output at the time of installation, the
ability of LED products to achieve satisfactory performance over time will contribute greatly to their
increased adoption. The capability for excellent color stability has already been demonstrated; as LED
technology continues to develop, there should be no need to compromise color stability for other
performance attributes.

26
6 References
[ANSI] American National Standards Institute. 2011. ANSI_ANSLG C78.377-2011. American National
Standard for electric lamps: specifications for the chromaticity of solid state lighting products. Rosslyn,
VA: National Electrical Manufacturers Association.
[ANSI] American National Standards Institute. 2004. ANSI C78.376-2001. American National Standard for
electric lamps: specifications for the chromaticity fluorescent lamps. Rosslyn, VA: National Electrical
Manufacturers Association.
[CIE] International Commission on Illumination. 2004. CIE 15-2004. Colorimetry. 3rd Edition. 72 p.
[DOE] U.S. Department of Energy. 2012. Demonstration of LED retrofit lamps. Host site: Smithsonian
American Art Museum, Washington, DC. PNNL-21476.
http://apps1.eere.energy.gov/buildings/publications/pdfs/ssl/2012_gateway_smithsonian.pdf
[EPA] Environmental Protection Agency ENERGY STAR Program. 2013. ENERGY STAR program
requirements, product specification for lamps (light bulbs); Eligibility criteria version 1.0. 25 p.
[EPA] Environmental Protection Agency ENERGY STAR Program. 2012. ENERGY STAR program
requirements for luminaires; Eligibility criteria version 1.1. 36 p.
http://www.energystar.gov/ia/partners/product_specs/program_reqs/Final_Luminaires_Program_Requ
irements.pdf
[IES] Illuminating Engineering Society. 2008. LM-80-08 IESNA approved method: measuring lumen
maintenance of LED light sources. 4 p.
[IES] Illuminating Engineering Society. 2011. TM-21-11 Projecting long term lumen maintenance of LED
light sources. 25 p.
MacAdam, DL. 1942. Visual sensitivities to color differences in daylight. Journal of the Optical Society of
America 32(5):247–274.
[NLPIP] National Lighting Product Information Program. 2004. What is color stability? Lighting Answers.
October 2004.
http://www.lrc.rpi.edu/programs/nlpip/lightinganswers/lightsources/whatisColorStability.asp
Wyszecki G, Stiles WS. 1982. Color science, concepts and methods: quantitative data and formulae. 2nd
ed. New York: Wiley; 968 p.

27
BUILDING TECHNOLOGIES OFFICE