Unit XIV.

Stellar Classification

675
 Classification of Stellar Spectra
As in many fundamental aspects of astronomy, the classification of stars is riddled with entertaining lore, having its
roots in the past (in this case the not-so-distant past; the current classification scheme is about a century old). The
classification system is constantly evolving; significant expansions occurred in the 1940s and in the late 1990s.
The detailed classification of stars became possible with the development of spectroscopy. Optical observations provided
brightnesses and colors of stars; with spectroscopy one could learn the compositions and physical conditions of stellar
atmospheres.
Newton passed sunlight through a prism and found that it consisted of light of many colors. In 1802 William Wollaston
(the discoverer of the elements palladium and rhodium) identified several dark lines in the solar spectrum; he suggested
that they marked the natural dividing lines between the colors (a band between blue and green, between green and
yellow, etc).
The great German optician Josef Fraunhofer (1787-1826) invented the spectroscope in 1814 and cataloged 574 absorp-
tion lines in the solar spectrum. (Remember that this investigation was performed without photographic materials!)
Fraunhofer deduced that the lines in the solar spectrum were caused by absorption by different elements, and succeeded
in identifying some of the material that resided in the solar photosphere (including iron). This was an astonishing
discovery! He was also able to observe the spectra of a number of bright stars and found that their spectra differed
from each other (e.g, not all stars possessed the same spectrum as the sun).
As spectrographs improved in quality and photographic plates became available later in the 19th century, extensive
analyses of stellar spectra became possible, confirming Fraunhofer’s finding that the stars possessed a wide range of
spectra. This discovery was not terribly surprising, as the easily visible range of the colors of stars demonstrated that
the general shape of stellar spectra could not be identical (although recall that the concept of blackbody radiation still
lay a half century in the future); what was unexpected was that the strength of absorption features, which were known
to be associated with particular elements and molecules, had enormous star-to-star variations! Does this behavior mean
that the stars have varying compositions?
Between 1860 and 1880 the Jesuit astronomer Angelo Secchi, based on his study of the spectra of more than 4,000
stars, developed the first stellar classification scheme. In 1866 he announced that there were three basic types of stars:
I) white and blue stars with strong hydrogen lines (e.g., Vega and Altair), II) yellow stars with weaker hydrogen lines
and some metallic features (the sun, Acturus, and Capella), and III) orange and red stars with complex “band” spectra
(Betelgeuse and Antares). Two years later he discovered the unusual class of red stars whose spectra are dominated by
carbon (Secchi Class IV), and in 1877 he added a fifth class: stars with emission lines in their spectra.

Angelo Secchi (1818-1878), an Italian Jesuit

priest, is recognized as the creator of the first
systematic stellar classification scheme, and is
credited with the first proof that the sun is indeed
a star. Unusually for European astronomers of
that era, he visited America; he taught at George-
town University, and developed a collaboration
with the Director of the Naval Observatory. He
also published drawings of the channels (which
were mistranslated into “canals”) on the surface
of Mars. He was the Director of the Roman
College Observatory from 1850 until his death.
Although known primarily for his many contri-
butions to astronomy, he also engaged in research
in meteorology and oceanography.

This work was a great advance in understanding the nature of stars, as it demonstrated that there were systematic
changes in the spectra. For example, since the basic physics of stars was not understood at the time, there would be
no reason to expect a correlation between color (temperature) and the appearance of various absorption features in the
spectrum. This information provided clues for future investigations. Why were there not any hydrogen features in red

676
 Angelo Secchi’s Stellar Classification Scheme (1866-1877)

I. White and blue stars with strong hydrogen lines (Vega, Altair)
II. Yellow stars with weaker hydrogen lines (sun, Acturus, Capella)
III. Orange and red stars with complex “band” spectra (Betelgeuse, Antares)
IV. Red stars whose spectra are dominated by carbon
V. Stars with emission lines in their spectra

(cooler stars)? Is the temperature of the star dependent upon the amount of hydrogen it contains (i.e., more hydrogen
makes a star hotter)?
Secchi made significant contributions to a number of fields in astronomy in addition to his classification scheme (comets,
meteors, absorption line formation in the solar atmosphere, the characteristics of the solar cornona, and planetary
astronomy). His detailed investigations of the spectra of the sun and stars produced the first definitive study demon-
strating that the sun was indeed a star!
E.C. Pickering and his group at Harvard began a spectroscopic survey in the 1890s that eventually led to the classification
of several hundred thousand stars. This program did not consist of several hundred thousand individual observations;
the data were collected using a technique called “objective prism spectroscopy”. Wide-angle field (hundreds of square
degrees) photographs of the sky were taken with a narrow prism placed in the beam; instead of recording images of
stars, this process produces low-resolution spectra of every object in the field.

This photograph of E.C. Pickering and his team

was taken at Harvard College Observatory on
13 May 1912. The back row, from left: Margaret
Harwood, Mollie O’Reilly, Pickering, Edith Gill,
Annie Jump Cannon, Evely Leland, Florence
Cushman, Marion Whyte, and Grace Brooks.
Front row, from left: Arville Walker, Johana
Mackie, Alta Carpenter, Mabel Gill, and Ida
Woods. This group, known as the “Harvard
Computers”, is largely responsible for the enor-
mous advances made in stellar classification in
the early 20th century. Pickering hired women
because their wages (25 to 50 cents per hour)
were significantly lower than those for men, thus
he could hire a larger staff given a fixed budget.
The story of the amazing accomplishments of this
group of women is presented in the book The
Glass Universe by Dava Sobel.

By this time improving astronomical instrumentation had allowed the Balmer lines to be identified in virtually all stellar
spectra, and the “Harvard Classification Scheme” used the strength of the Balmer lines as the primary classification
criterion. The stars with the strongest Balmer lines were called A stars, stars with slightly weaker Balmer lines B stars,
and so on down to P stars, which do not possess any hydrogen features.
Why are stellar spectra so different? Not only do the Balmer lines vary dramatically in strength, but the absorption
features of all the elements change between the different stellar types. The obvious explanation is that the composition
of stars had a large variation; some stars were predominately hydrogen, others were virtually bereft of the element! (In
this interpretation A stars were the most hydrogren rich, P stars the most hydrogen poor.)
While the Balmer-line based scheme produced a smooth variation of hydrogen line strength with spectral type, it did
not produce satisfactory curves for the strengths of many other elements. For example, some helium lines were strongest

677
in “O” stars (virtually no hydrogen), were essentially absent in solar-like stars (class “G”), but were moderately strong
in “B” stars! In addition, A and O stars, at opposite ends of this classification scheme, were both blue, while the reddest
stars were near class M! Annie Jump Cannon, one of Pickering’s assistants, examined the dependence of absorption
features on spectral class; her goal was to see if it was possible to order the spectral types such that the strengths of
all features were a continuous function of spectral type.
Cannon was able to achieve her goal and modified the original classification scheme; her system was adopted as the
international standard in 1922 and is still in use today (and will probably be in use for centuries). Cannon’s major
insight was that the Harvard system was basically produced by differing stellar temperatures and was not necessarily a
composition indicator. By examining the absorption features of all elements and requiring that the relative strengths
of the lines were a smoothly varying function of temperature (the temperature of maximum strength of a line depended
on the type of line), she was able to organize the spectral classes to reflect the temperature sequence. She published
her first spectral classifications in 1901.

Annie Jump Cannon (1863-1941) of the Harvard

Observatory created the basis for the stellar
classification scheme that has been used for a
century. Using the objective prism plates that
had been obtained as part of a wide field spec-
troscopic survey, she was able to organize stellar
spectra into a temperature sequence, and her
OBAFGKM classification scheme was adopted
by the astronomical community. The American
Astronomical Society annually awards the Annie
Jump Cannon Prize to recognize the accom-
plishments of a young North American female
astronomer.

Cannon revised the spectral classes to the following system:

OBAFGKM

The hottest (bluest) stars are the O stars, the coolest (reddest) are the M stars. Stars near the front of this sequence
are called early-type stars; stars near the end are called late-type stars (more on this point later). The hottest and the
coolest stars both have extremely weak hydrogen features, while the strongest Balmer lines occur near the middle of
the temperature scale! (The strengths of the Balmer lines, however, do change in a smooth, systematic manner along
the spectral sequence.)
Cannon removed a number of spectral classes (e.g., there are no C, D, E, H, etc stars) but refined the classification by
creating ten subclasses for each class. For example, instead of simply calling a star a “B” star, she divided this group
into B0, B1, B2, ... B8, and B9 stars, with A0 following B9. If an object has a classification of B0 (or B1), it would be
called an early-B star; B stars with large subclass numbers are referred to as late-B stars.
While Cannon’s system had a sound physical basis, it was hard to remember; generations of astronomers (including
this one) were trained with the expression “Oh be a fine girl kiss me.” This expression is now viewed as reactionary,
so you are on your own as to how to memorize the sequence. Cannon’s achievements are honored by the American
Astronomical Society’s Annie Jump Cannon Prize, given annually to an outstanding achievement by a young North
American female astronomer. In 2007 the award was given to Ann Hornschemeier, who completed her PhD at Penn
State in 2003 with Professor Niel Brandt; Penn State Professor Rebekah Dawson received the prize in 2017.
Cannon’s stellar classifications were included in the Henry Draper Catalog; the original complete edition contained
225,300 stars and was published in 1924. (Draper was a wealthy amateur astronomer whose widow funded the telescope

678
used in the survey.) Many bright stars are still known by their “HD” number; for example, Betelgeuse is HD 39801.
Some examples of stellar types of well-known stars are given in the accompanying table; the figure displays the spectra
of a series of stars from type O through type M.

Spectral Types of Bright Stars

Star Spectral HD
Name Type Number

Spica B1 116658
Rigel B8 34085
Vega A0 172167
Sirius A1 48195
Procyon F5 61421
Sun G2 ——
Arcturus K2 124897
Antares M1 148878
Betelgeuse M2 39801

The explanation of the large variety of features seen in stellar spectra is not composition variations; the changes are
due to the differing physical conditions (temperature, pressure, gravity) in the atmospheres of the stars. It was not
until 1925 that Cecilia Payne Gaposchkin was able to perform the first accurate estimates of the compositions of stars
(later published in her PhD thesis!). Her results indicated that the compositions of stars are remarkably similar; there
are some important variations, but to first order most of the mass is hydrogen and the vast majority of the remainder
is helium, which contradicted the accepted view at that time that the sun and the Earth had similar compositions.
(Indeed, one of the leading American astronomers, upon reading this work, declared that it could not possibly be
correct. Several years later additional observations revealed that Gaposchkin’s interpretation was accurate; the leading
astronomer pronounced that stars had similar abundances, and today he is credited in many circles with making the
initial discovery!)

In her 1927 PhD thesis Cecilia Payne Gaposchkin

(1900-1979) demonstrated that all stars had
essentially the same chemical compositions,
formed primarily of hydrogen and all but a tiny
fraction of the remainder was helium. The radi-
cally different appearances of the strengths of the
features in stars were not due to chemical abun-
dance variations but to different surface tempera-
tures. She was the first winner of the Annie Jump
Cannon Award (1934) of the American Astro-
nomical Society.

Before proceeding, I should say a few words about compositions of stars. To astronomers, there are basically only three
elements: hydrogen, helium, and “metals”. The term metals refers to everything that is heavier than helium; oxygen,
mercury, and uranium are all “astronomical” metals. You can see that the astronomers’ periodic table is considerably
simpler than that of chemists!
When describing compositions, the quantity used by astronomers is the mass fraction due to a given element. The
mass fraction due to hydrogen is symbolized by X, that of helium Y, and that for metals Z. With this definition is it
clear that
X + Y + Z = 1

679
 F7
O5
0

G0
O9

G2
B3
5

G6
B6
Normalized AB magnitude

K0
A1

K5
A5
10

M1
A8

M5
F3
15

4000 5000 6000 7000 4000 5000 6000 7000

Wavelength (A) Wavelength (A)

Optical spectra of a set of stars with spectral types of O5 to M5. Type O are the hottest stars; the optical band is
on the Rayleigh-Jeans tail of the spectrum (the peak is shortward of 3000 Å). As one moves down the left-hand panel,
the Balmer lines grow in strength, reaching a maximum in the early A stars, then these features gradually diminish.
In A stars the overlapping Balmer lines near the Balmer break (3636 Å) absorb nearly all of the radiation in the near
ultraviolet; for these stars the U − B color index is red, but the B − V value indicates a hot star! The right panel
displays a set of cool stars (the sun has a spectral type of G2); the hydrogen features disappear and the metal lines
(particularly the H and K lines of Ca II near 3950 Å in G/K stars and molecular bands in M stars) begin to dominate
the spectrum as one moves down the column. The absorption feature at 6900 Å (present in all the spectra) is due to
the Earth’s atmosphere (the Fraunhofer B line).

A typical star of solar composition has

X = 0.70 Y = 0.28 Z = 0.02
Remember that these quantities are mass fractions; if you are talking about number fractions (e.g., if you pick an atom
at random, what is the probability that it is a hydrogen atom?), then hydrogen is even more dominant; about 90% of

680
 The astronomical periodic table of the elements.
The number associated with each element is its
mass fraction (normalized to 10,000). Hydrogen
and helium constitute approximately 98% of the
mass in the universe; oxygen (1%) and carbon
(0.5%) are the bulk of the remaining mass.

the atoms in such a star would be hydrogen, and much, much fewer than 2% of the particles would be heavier than
helium.
Stars do vary in composition, but the main difference lies in the fraction of metals. There are stars that are extremely
metal-poor; their values of Z are on the order of 0.001 (i.e., 0.1% of the mass is in elements heavier than helium) or less.

To obtain a good understanding of stellar atmospheres requires much more time that we have available in this class;
entire graduate-level courses are devoted to this single topic. I will quickly introduce the basic concepts in the next
lecture or two to provide an overview of the subject and the important physical processes, but won’t delve into as much
detail as is presented in the textbook.
Equivalent Widths
To develop a classification scheme based on the spectra of hundreds of thousands of stars, one must create a systematic
measurement technique to characterize the strengths of absorption features. The ideal system would assign a single
value to each line; this measurement should be simple, consistent, objective, and quantitative. In addition, since we
are interested in the intrinsic properties of the star, the quantity should be independent of distance; e.g., two identical
stars at different distances from the Earth should produce the same value of the absorption line strength.
The obvious measurement to adopt to characterize line strengths is the amount of “missing flux”. What is the total
amount of flux removed? If one defines a “continuum” level (fλ (λ)[cont]), the missing flux in the case of a single
absorption feature in the spectrum is
Z λ2
∆f = [fλ (λ)[cont] − fλ (λ)] dλ
λ1

The wavelength limits of integration are set to fully cover the region affected by the absorption feature.
This quantity is, however, not that useful for classification, since ∆f depends on the brightness of the star, thus the
measured value depends on the location of the observer. For example, consider two identical stars, one located at a
distance of 1 pc, the other at 10 pc. The strength of the absorption line in the more distant star, characterized by ∆f ,
would be only 1% of that of the nearer star. This measured difference for physically identical objects is clearly not a
desirable trait; we want a measurement that will produce the same value for a star independent of its distance, i.e., a
“normalized” absorption that records what fraction of the light is absorbed by a feature.
Astronomers have adopted the concept of Equivalent Width (W ) to indicate the strength of a spectral line. This
quantity is the width of a “rectangular”-shaped, completely black line that removes the same amount of flux as the
observed feature (see accompanying figure).
The width of the equivalent rectangular (blue vertical sizes in the figure) region is called the equivalent width of the
line. The units of equivalent width are length. For lines occuring in the optical, the equivalent width is usually given
in Angstroms (not cm or m); strong lines have equivalent widths of several Angstroms. Equivalent width is also defined
for emission lines; the equivalent width of an emission line is the width of the continuum region that contains the same
amount of flux as the emission line. The equivalent widths for emission and absorption lines have opposite signs.

681
 1.00
The definition of the equivalent width of a
line. The spectrum is first normalized to the
continuum level. The area removed by the line
(a Gaussian profile in this example) is equal to
0.75

the rectangular area enclosed by the blue verticle

Normalized Flux Density

lines. The equivalent width (W ) is the amount

of wavelength (30 Å in this diagram) covered
0.50

by the rectangle; i.e., this is the amount of

spectrum that would be completely “black” by
a rectangular-shaped feature that removed the
same amount of flux as the observed line. The
0.25

equivalent width is independent of the distance to

the star, the resolution of the spectrograph, and
W the fraction of the flux that is recorded by the
0.00

instrument. The equivalent width of an emission

line is defined in a similar fashion.
6400 6500 6600
Wavelength (A)

The formal definition of the equivalent width is similar to the previous expression for ∆f , but one normalizes the
observed flux density with the continuum. This approach has the desired feature that identical stars at different
distances from the Earth will yield identical measurements for the equivalent widths of lines.
Equivalent width is defined as (where fcont is fλ (λ)[cont])
Z λ2 Z λ2
fcont − fλ (λ) fλ (λ)
W = dλ = 1 − dλ
λ1
fcont λ1
fcont

Equivalent Width

λ2 λ2
fcont − fλ (λ) fλ (λ)
Z Z
W = dλ = 1 − dλ
λ1 fcont λ1 fcont

This definition has several attractive features

• Independent of the brightness (distance) of the star
• Independent of what fraction of the light is measured
• Independent of the resolution of the spectrograph
W has the units of length (typically Å)

This definition has the desired distance-independent feature, but it also is attractive because of its practical aspects;
since equivalent width does not depend on the fraction of total light captured or the spectral resolution, it is easy to
compare results from different observing conditions and spectrographs. For example, if one had proceeded with the
simple “absolute value of the flux removed” definition, one would always have to be certain that 100% of the light was
captured, and accurate atmospheric transmission corrections made for each observation.
As an example, consider a star with an absorption feature that removes 1.0 × 10−15 erg s−1 cm−2 from a spectrum
that has a continuum level of 1.0 × 10−8 erg s−1 cm−2 cm−1 . This line has an equivalent width of 10−7 cm, or 10 Å.
If the star’s distance from Earth is increased by a factor of two, both the continuum flux density and the flux removed
by the line will drop by a factor of four; thus the equivalent width of the line remains the same. Also consider

682
the complications that arise in the absolute flux measurement if a small cloud passes in front of the star during the
measurement and reduces the transmitted flux by 10% (or your telescope does not track properly and the star is outside
of the spectrograph aperture 10% of the time). If this event occurs, your absolute measurement will be incorrect by 10%,
but the value of the equivalent width will be unaffected! In addition, if this star was observed with spectrographs of
differing spectral resolution or quantum efficiency, the measured value of the equivalent width would be the same (with
the caveat that lower resolution and lower quantum efficiency would generally reduce the signal-to-noise ratio of the
measurement in a given exposure time). The line profile in the lower-resolution spectrograph would generally be wider
and shallower than in the higher-resolution instrument, but the equivalent width value would be the same.

The amount of flux removed from the beam is related to the number of absorbers (actually absorbers plus scatters)
present. When the optical depth of the material is much less than one, the amount of absorption (e.g., the equivalent
width) is proportional to the number of absorbers in the beam because there is no “self shadowing” by the absorbers.
For example, if there is one hydrogen atom in the beam, adding another such atom with double the absorption (and
hence the equivalent width). Once the optical depth approaches and exceeds one, however, the equivalent width is
no longer proportional to the number of absorbers. This phenomenon is easy to visualize; once the center of the line
becomes essentially “black” (no transmitted photons), piling more absorbers into the line will not remove additional
flux (as there is none to remove!). The quantity that is proportional to the number of absorbers is the optical depth
(recall the radiative transfer equations); when τ is much less than one, the absorbed flux is proportional to τ (and
therefore to the number of absorbers).

The Boltzmann and Saha Equations

What determines the strength of an absorption line as a function of temperature? Clearly the most important quantity
that influences absorption line properties is the number of atoms present that are in the proper configuration to absorb
a photon of that energy. If a star is bereft of iron, its spectrum cannot possess any iron lines!

Before being buried with a blizzard of equations, let us examine what can be deduced about line strengths from our
basic understanding of blackbody radiation, ideal gases, and atomic structure. Throughout this section we will assume
that we are dealing with material in thermal equilibrium, i.e., a single temperature characterizes the radiation field and
the kinetic motion of matter. (This condition does not always apply in astrophysical situations, but we will not discuss
this complicated topic in this class.)

Imagine that you are a completely isolated hydrogen atom (this example works for any atom/molecule): you never
encounter any other particle or photon. If you are in the ground state, the electron cannot ever become excited. If the
electron resides in an excited state, it cannot ever move to a higher state, but it can move to a lower state by emitting
a photon. In the majority of cases the de-excitation occurs on time scales that are a tiny fraction of a second.

At extremely low temperatures (near absolute zero), the radiation field is weak and the particles are moving very, very
slowly, and we have a similar situation to that described in the previous paragraph. All of the electrons will be in
their ground state, since neither collisons or photon absorption can cause significant numbers of electrons to exist at
excited states. The only observed absorption features will be those that originate in the ground state. (In general if
absorption occurs, then the atom will quickly return to the ground state before it can absorb a photon for a higher level
feature.) As the temperature increases, the kinetic energy available from collisions (and the frequency of collisions)
rises, as does the number of photons that can, when absorbed, excite the electrons; one would expect to see absorption
features that arise from exited levels (e.g., the Balmer series) to appear. For the case of hydrogen, the first excited
state presents quite a hurdle. The temperature associated (kT ) with an energy of one electron volt is 11,600 K, and
the energy required to move to the first excited state is 10.19 eV!

As one continues to increase the temperature, exitation mechanisms (both collisionial and radiative) allow electrons to
be boosted into ever higher levels of excitation. At some point the kinetic and photon energies reach the point where
the atom can be ionized, thus entirely removing the “discrete” absorption produced by bound-bound transitions! (This
behavior provides an insight into absorption line strengths: in the hottest stars [type O], the temperatures may be so
large that all of the hydrogen is ionized, and therefore no Balmer lines would be formed, even if the star was entirely
componsed of the element!)

Throughout this class I prefer to derive the physical laws we use; I don’t want to act as Moses coming down from the
mountain and announce “This is the way it is.” Unfortunately, the background necessary to derive the strengths of
absorption lines is beyond the scope of this course, so “This is the way it is.”

683
I’m going to state without derivation that in a system in thermal equilibrium (a single value of the temperature of the
system characterizes all energy levels) the ratio of the number of atoms in levels A and B is given by the Boltzmann
Excitation Equilibrium Equation:
EB
g e− kT EA −EB

∆EAB


NB g g
= B E = B e kT
= B e kT
where ∆EAB = EA − EB
NA A gA gA
gA e− kT

In this formula gB and gA are the quantum mechanical statistical weights of each of the levels (dimensionless numbers
of order unity), EA and EB are the energy levels of states A and B, and T is the temperature. The statistical weights
are basically the number of different quantum states that have a given energy (i.e., takes into account degeneracy).

Ludwig Boltzmann (1844-1906) was the founder

of the field of statistical mechanics; we have
encountered two of his fundamental constants (k
and σ) previously in the course. He was on the
faculty at a number of Austrian universities. He
suffered from severe depression, and hung himself
while on a summer vacation. His formula for
entropy, S = k log W , is engraved on his tomb-
stone.

Notice the familar form of the exponential: an energy (in this case an energy difference) divided by the characteristic
energy of the system (kT ). (This factor also appears in the Maxwellian Distribution.) The “interesting” temperature
region is when the system temperature is comparable to the energy difference; at temperatures far below kT , all the
particles are in the lower energy state; once the temperature considerably exceeds kT , the ratio becomes relatively
insensitive to the temperature (the levels are populated at the ratio of the statistical weights in the case of thermal
equilibrium).

Boltzmann Excitation Equilibrium Equation

When material is in thermal equilibrium, the ratio of the number of particles in state B to the number
of particles in state A is
EB
NB gB e− kT
= EA
NA gA e− kT

EA and EB are the energies of states A and B. The quantities gA and gB are the quantum mechanical
statistical weights of states A and B; these weights are dimensionless quantities whose values are of
order unity.

One often encounters the following form of the Boltzman equation:

NB gB gB
EA −EB

∆EAB


= e kT
= e kT
(∆EAB = EA − EB )
NA gA gA

Let us examine the case where EB is larger than EA (such as the case for the hydrogren atom when nB > nA ). At
low temperatures the ratio is essentially zero because of the exponential term. As the temperature rises, the ratio

684
monotonically increases; at infinite temperature the ratio of the two states becomes the value of the ratio of the
statistical weights. (If the two energies are equal, the ratio of the two states is always equal to the ratio of the statistical
weights in the case of thermal equilibrium.)
We can now apply the fundamental aspects of line formation to the observed behavior of the fiducial feature of the
Harvard classification scheme, the Balmer lines, as a function of temperature. Balmer absorption occurs when a
hydrogen atom with an electron in the n = 2 level (the first excited state) encounters a photon with the energy required
to pump the electron to a higher energy level. The accompanying table lists the properties of the first several lines of
the Balmer series.

Properties of the Balmer Lines

Energy
Line λ (Å) nA nB (eV)

Hα 6563 2 3 1.89
Hβ 4861 2 4 2.56
Hγ 4340 2 5 2.86
Hδ 4101 2 6 3.03
Hǫ 3970 2 7 3.13
Hζ 3889 2 8 3.20
Hη 3835 2 9 3.24
Limit 3646 2 ∞ 3.41

Clearly a key parameter in determining the strength of the Balmer lines is the fraction of hydrogen atoms in the n = 2
state as a function of temperature. If there are no atoms with electrons in the first excited state, then obviously no
Balmer absorption lines can be produced! From our previous discussion, it is obvious why Balmer lines are weak at
low temperatures: Balmer lines are not ground-state transitions, so at low temperatures there cannot be any hydrogen
atoms in the first excited state!
Let us transform Boltzmann’s constant for easy use in atoms, where the energy differences are typically measured in
electron volts (1 eV = 1.6022 × 10−12 erg).
The values of Boltzmann’s constant in cgs, mks, and “atomic” units are

k = 1.3806 × 10−16 erg K−1 = 1.3806 × 10−23 J K−1 = 8.6173 × 10−5 eV K−1

Thus to have kT equal to one electron volt (the energy of a photon with a wavelength of 1.24 µ), one must have a
temperature of ≈ 11,600 K.
What temperatures are favorable for the production of Balmer lines? Imagine that the classroom was filled with
hydrogen gas and the gas was in thermal equilibrium with the environment (i.e., a temperature of around 300 K).
The first excited state (n = 2) has an energy level that is 10.20 eV higher than the ground state; therefore to have
an appreciable number of atoms in the first excited state one must have the value of kT on the order of 10 eV, or a
temperature of approximately 100,000 K!! The characteristic energy (thermal photon and kinetic) of a system with a
temperature of 300 K is only 0.026 eV, which is over two orders of magnitude too small to place an electron in hydrogen
atom into the first excited state!
Let us now examine this issue in a quantitative manner. For hydrogen atoms, the expression for the statistical weights
is simply
gn = 2n2

Aside for those who have had introductory quantum mechanics: It is relatively easy to derive the statistical weights for
the hydrogen atom. For a given state n, the maximum value of the angular momentum quantum number (l) is n − 1; a
state l has 2l + 1 levels (the “z-axis” projection runs from −l, (−l + 1), ..., −1, 0, 1, ....(l − 1), l). The n = 1 state has only
the 1s orbital, which has one possible angular momentum (l = 0) and two possible electron spin states, for a statistical
weight of two. The n = 2 state has two orbitals, 2s and 2p, which have one and three possible angular momentum states,
respectively. Taking into account electron spin (two orientations), the statistical weight is 2(1 + 3) = 8. Extending this

685
analysis to higher levels, the total number of angular momentum states for quantum number n is the sum of the
sequence of odd numbers (2i − 1) ending at i = n; this sum is simply n2 . The factor of two arises from the two possible
spin states. (The calculations of the g-factors for multielectron atoms can be exceedingly complex.)
The number ratio of the first excited state (n = 2) to the ground state (n = 1) in H I, given by the Boltzmann equation,
is (expressed using the electron volt energy scale)
 
N2 g E1 − E2 8 −13.605 + 3.401 118, 000
   
= 2 exp = exp = 4 exp −
N1 g1 kT 2 (8.62 × 10−5 ) T T

Obviously the temperature must be above 500 K before there can be any atoms in the n = 2 state (There are around
1080 atoms in the universe!). Even at infinite temperature there are only four times as many atoms in the second as
the ground state. (Caveat: at some point the temperature will be sufficiently high that the hydrogen will be ionized;
this aspect of line formation will be discussed in a few pages.)
The number in the first excited state equals the number in the ground state when

118, 000 118, 000 118, 000

 
1 = 4 exp − ⇒ ln 4 = ⇒ T = = 85, 100 K
T T ln 4

This result does not mean, however, that half of the hydrogen atoms are in the first excited state at this temperature,
for there are the higher order states to consider as well (n = 3, 4, etc). We can use the Boltzmann equation to find the
ratios for all of the adjacent states in the atom.
The number ratio of the second to first excited state as a function of temperature is
 
N3 18 −3.401 + 1.512 21, 900
 
= exp = 2.25 exp −
N2 8 (8.62 × 10−5 ) T T

At 85,100 K, this ratio is 1.74! (The ratio is around one at 27,000 K). And there are still the large number of higher
states to consider! This result means that it is impossible to have half of the hydrogen atoms be in the first excited
state for a system in thermal equilibrium; long before the number in the first excited state equals the ground state, the
number in n = 3 exceeds the number in n = 2! (And, of course, there are additional higher energy levels available to
the electrons.)
One can see that the number of hydrogen atoms in the first excited state is strongly dependent on the temperature;
to calculate the exact value, a large number of levels must usually be considered. How does one actually calculate the
fraction in a given state? The fraction of atoms in state n is

Nn Nn
=
Ntot N1 + N2 + N3 + N4 + . . .

The right hand side of the equation is not in a suitable form to employ the Boltzmann equation. Dividing the numerator
and denominator of the right-hand expression by the number of particles in the ground state (N1 ) yields

Nn
Nn N1
=
Ntot N2
1+ N + N + N
N3 4
N1 + . . .
1 1

The ratios can all be calculated from Boltzmann’s equation. Boltzmann’s equation is not restricted to adjacent energy
N7
levels; one can calcuate, for example, N directly. (One could have chosen to divide by the number of atoms at any
2
given level and produced a manageable equation, but by convention the ground state is chosen to be the normalizing
configuration.)
Astronomers often represent the sum of the ratios (up to a constant factor) given in denominator of the previous
equation by a quantity called the partition function. Each element (denoted by the subscript e) has a partition
function, ue (T ), which is a function of temperature and is defined as

j=∞
X Ej −E1
ue (T ) = gj e− kT

j=1

686
Although the notation used for the the partitition function implies that this quantity is a function only of temperature,
each element has its own distinct partition function (differing statistical weights and energy levels; don’t forget the “e”
subscript!).
The partition function can be tabulated for each element as a function of temperature. Given this definition, the
fraction of atoms in a given state can be written as
En −E1
Nn gn e− kT
=
Ntot ue (T )

Partition Function

The partition function is used to calculate the fraction of atoms in a given state.
Nn
Nn Nn N1
= = N2 N3
Ntot N1 + N2 + N3 + N4 + . . . 1+ N1
+ N1
+N4
N1
+ ...

The ratios can all be calculated from Boltzmann’s equation.

The definition of the partition function and the fraction of atoms in a given state are
j=∞ En −E1
X
−
Ej −E1 Nn gn e− kT
ue (T ) = gj e kT and =
j=1
Ntot ue (T )

In essentially all situations, the partition function diverges, so one must terminate the summation at
energy levels that cannot be supported in the material (e.g., the size of the orbital must be significantly
smaller than the average separation of atoms).

Given your knowledge of the hydrogen atom, a little thought reveals the most unpleasant fact that the partition function
diverges at essentially any temperature! The partition function contains an infinite number of terms; as one approaches
large n, the exponential factor approaches a constant (as n approaches infinity, the energy difference approaches a
constant value, the ionization energy; in the case of hydrogen, 13.605 eV), whereas the statistical weights of the levels
grow without bound (n2 ). In the real world, however, all levels cannot be populated; the size of the orbit (and hence the
limit on the quantum number) cannot be larger than the average spacings between atoms! For example, if n ≈ 10, 000
in a hydrogen atom, the electron is about one centimeter from the nucleus! In actual physical situations, one must
determine at what level to truncate the partition function.
At low temperatures, the value of the partition function approaches the statistical weight of the ground state (all the
useable terms in the sum can be made vanishingly small compared to the first term, which is simply g1 ). In this case
the previous equation reduces to the Boltmann equation for the ratio of state n to the ground state, since almost all of
the material is indeed in the ground state.
Of course real life is yet more complicated than this picture; if the temperature becomes sufficiently large, there will be a
population of photons capable of ionizing the atom (in the case of hydrogen, photons with hν > 13.6 eV or λ < 912 Å).
In addition, the energy exchange in collisions between particles can result in ionization. At this point the hydrogen
gas becomes a plasma, composed of free electrons and protons, and Balmer absorption lines (indeed, all bound-bound
absorptions) cannot be created.
We have discussed the Boltzmann equation in the context of the relative numbers of atoms in various bound-bound
states, but the formula also applies in the “particle physics domain” where one has particles that can transform
into different types of particles. The classic example is the neutron-proton ratio (which is an important quantity
when discussing the conditions of the early universe). Beta decay (discussed later this semester in the context of the
mysterious class of particles called neutrinos) occurs when a neutron changes into a proton, electron, and neutrino. If
one has a free neutron (i.e., not confined to a nucleus), it will beta decay with a half-life of 617 s. In a system that is in

687
thermal equilibrium, the ratio of neutrons to protons can be calculated from the Boltzman equation, with the protons
being the lower energy state.
There is an equation, derived in section 8.1 of the Carroll and Ostlie, that yields the ratios of the various ionization
states for a specific element. This formula is called the Saha equation, and provides the ratio of the number of
particles in the (i + 1)th ionization state to the number of particles in the ith state:

Ni+1
 32
2kT ui+1 2πme kT
 χi
= e− kT
Ni Pe ui h2

In the formula Pe is the electron pressure, ui+1 and ui are the partition functions for the two ionization states (not only
does each element have a partition function, but every ionization state of each element has its own partition function!),
and χi is the ionization energy from the ground level of the ith state to the ground level of the (i + 1)th state. The
Saha equation is similar in form to the Boltzmann Equation (providing the ratio of the number of particles in different
states, including the appearance of the the familar exponential term with the ratio of the required change in energy to
the “system energy” kT ).

A 1921 photograph of Meghnad Saha (1893-

1956), the physicist who solved the problem of
calculating the ionization state of atoms given
the surrounding physical conditions; his solution
is known as the Saha equation. He was born in
poverty in what is now Bangladesh, and rose to
become a Fellow of the Royal Society (at age 34)
and Dean of the Faculty of Science at Calcutta
University.

We will not delve into details of ionization here; I simply wish to provide a sense of what is needed to calculate
the properties of a stellar atmosphere. The ratio of the number of atoms in two ionization states depends upon the
ionization energy from one state to the next (if there is a large ionization energy, it is more difficult to ionize, and larger
temperatures are required to equalize the numbers in the two states), temperature-sensitive functions that are specific
to a given transition, the temperature (higher temperature, more ionization), and, surprisingly, the pressure due to free
electrons.
The last factor enters because a free electron is always in search of a home (a nucleus); if there is a high density of free
electrons, the lifetime of an ionized atom is shorter than if there were fewer free electrons. For example, imagine that
we have two boxes of different sizes, each containing a the same number of hydrogen atoms at a certain temperature.
The gas in the box with the larger volume (lower gas density and hence lower pressure) would be more highly ionized
than the gas in the smaller box, because free electrons in the smaller box would encounter nuclei more often than
the large-box electrons. In addition, the faster the electrons move the quicker they will encounter a nucleus; thus the
important quantity in the Saha equation is the electron pressure (proportional to the density times the velocity). (One
can, of course, replace the electron pressure in the Saha Equation with the expression ne kT .)
Armed with this knowledge, we can now understand the behavior of the Balmer lines with temperature. The strength of
the absorption depends on the number of neutral hydrogen atoms in the first excited (n = 2) state. At low temperatures,
all the atoms are neutral, but all of the electrons are in the ground state. As the temperature increases, the number of
n = 2 atoms increases, but eventually 1) electrons head for even higher levels and 2) atoms begin to be ionized. The
accompanying figure shows, for the case of an electron pressure of 10 dynes cm−2 (typical in the solar photosphere),
the fraction of ionized hydrogen as a function of temperature. At this electron pressure, there is virtually no ionization
until the temperature reaches around 7000 K, and the ionization is essentially complete by 10,000 K.

688
 Saha Ionization Equation

When material is in thermal equilibrium, the ratio of the number of particles in ionization state i + 1
to the number of particles in ionization state i is
  32
Ni+1 2kT ui+1 2πme kT χi
= e− kT
Ni P e ui h2

where Pe is the electron pressure, ui+1 and ui are the partition functions for the two ionization states,
and χi is the ionization energy from the ground level of the ith state to the ground level of the (i+1)th
state.
Note the familar exponential term with the ratio of the required change in energy to the “system
energy” (kT ).
1.00
0.75

The fraction of hydrogen that is in the ionized

state as a function of temperature for an electron
Ionized Fraction

pressure of 10 dynes cm−2 , which is typical for

0.50

the solar photosphere. The ionization occurs over

a relatively small temperataure range; hydrogen
is essentially entirely neutral at this electron pres-
sure at 7000 K; by 10,000 K, the material is
0.25

almost totally ionized.

0.00

2500 5000 7500 10000

Temperature (K)
-2
Hydrogen Pe = 1.00E+01 dynes cm

The following figure charts the rise and fall of the fraction of hydrogren atoms in first excited state as a function
of temperature (for the conditions in the solar photosphere). As was seen in the previous ionization curve, at low
temperatures virtually all the atoms are in the ground state. At temperatures around 5000 K, excited levels begin to
be populated, the first one being the n = 2 state. As the temperature rises above this point, an increasing number of
atoms are excited above the n = 2 level (recall that at a temperature of 27,000 K the number of atoms in the n =2
and n = 3 states are equal), so the number of atoms in the first excited state declines. The “Balmer fraction” reaches
a maximum around 8500 K, and declines at larger temperatures due to competition with higher levels and (mainly)
ionization.
Even at the peak, however, the fraction of atoms in the first excited state never far exceeds one part in a million!! The
difficulty of populating this state arises because the energy difference between the ground state and the n = 2 level in
hydrogen is 75% of the energy required to ionize the atom. By the time the system temperature has reached the energy
level required to significantly populate the first excited state, the system is almost to the ionization requirement. If
we lived in a universe where E2 = −11 eV, then the first excited state could be reached at temperatures where the
ionization fraction would be vanishingly small.
The strength of the Balmer lines is reflected in this curve. In cool stars, the lines are weak because there are few atoms
available for Balmer absorption. At temperatures around 8,000 K to 10,000 K, the fraction of hydrogen atoms in the
n = 2 state reaches its maximum, and the Balmer lines are strongest. As one moves to higher temperatures, the number

689
 The fraction of hydrogen that is in the first
excited state (n = 2) as a function of temper-
-6

1.0
Fraction of Hydrogen [n = 2/Total] x 10

ature (again, for an electron pressure of

10 dynes cm−2 ). At this electron pressure only
about one hydrogen atom in a million can be
placed in the first excited state and thus produce
Balmer absorption. At low temperature the
0.5

Balmer lines are weak because the vast majority

of electrons are in the ground state; at high
temperatures the Balmer lines disappear because
no neutral atoms remain!
0.0

0 10000 20000 30000

Temperature (K)
-2
Hydrogen Pe = 1.00E+01 dynes cm

of atoms that can produce Balmer lines decreases, so the Balmer line strengths decrease in the hottest stars.
If the electron pressure in the solar atmosphere increased, the peak of the curve in the figure would shift towards
higher temperatures and reach a larger value for the fraction n = 2 atoms. The larger electron pressure would suppress
ionization at a given temperature, so there would be more neutral atoms to take advantage of the rising number of
atoms in the first excited state (via the Boltzmann equation) with temperature.
This phenomenon of the “rise and fall” of lines is characteristic of all neutral non-ground state atomic transitions. The
strengths of all non-ground state transitions rise and fall with temperature because of the changing population of the
various energy levels and the change in the ionization state. (Ground state transitions monotonically decrease with
increasing temperature, because at zero temperature all atoms are in the ground state. An important caveat is that the
ground level population of ionized states is not maximized at zero temperature because, of course, at that temperature
nothing is ionized!)

This diagram displays the rise and fall of

strengths of absorption features with tempera-
ture. For non-ground state transitions (e.g., the
Balmer lines), as one moves from lower to higher
temperatures (right to left in the diagram), the
line strength initially increases, as the initial level
of the transition begins to be populated. The
strength is maximized when the rate of popula-
tion gain from lower states is balanced by the
loss of electrons to higher states and ioniza-
tion (e.g., at approximately 10,000 K, or spec-
tral type A0, for Balmer lines). The strength
of ground state transitions steadily decline with
increasing temperature.

Examining this figure reveals how Annie Jump Cannon was able to determine a relative temperature sequence without
any knowledge of quantum mechanics, energy levels, etc. Recall that the equivalent width of an absorption line is related
to the number of atoms in the state that can give rise to the absorption (as mentioned previously, in some limited but
interesting situations, the equivalent width is directly proportional to the number of absorbers; double the number of
absorbers and the equivalent width increases by a factor of two). She had a dataset consisting of a large number of
stars, each with an assigned spectral type and equivalent widths of absorption lines appearing in the spectrum. Cannon
demonstrated that it was possible to organize the spectral types in a sequence that produced smooth, continuous curves
for all of the line strengths.
The presence of absorption lines causes the spectral energy distribution of stars to deviate from that of blackbody
radiation, sometimes significantly. There are conditions where the lines are so strong and so closely spaced that the

690
spectrum in that region is strongly suppressed; this phenomenon has been given the name line blanketing. When line
blanketing occurs, the energy is redistributed in the spectrum (in optically thick media in radiative equilibrium, lines
do not remove energy from the beam; the energy still emerges but at wavelengths where the opacity is smaller). This
effect can introduce significant changes in the form of the color index-temperature relation.
The classic example of line blanketing in stars is the effect of the high order Balmer lines on the spectrum of A stars.
The lines near the Balmer limit (3646 Å) overlap (there are, in theory, an infinite number of Balmer lines, although the
quantum number is limited by the physical separation of the atoms!), producing a dramatic drop in the spectrum near
the series limit. In addition, any photon with wavelengths shorter than the Balmer limit has sufficient energy to ionize
a hydrogen atom in the first excited state, so this radiation is also highly susceptible to absorpton.
If one forms the (B − V ) color index for these stars, one finds a monotonic relationship with spectral class (i.e.,
temperature): the hotter the star, the bluer the color index. The same is not true for the (U − B) color index. At low
temperatures, the index is large and positive, as expected, but as one reaches the A stars the line blanketing due to the
Balmer lines becomes so strong that the (U − B) color index for A stars is larger (redder) than that of F stars!

M7
M6
M5
M4
1.5

M3
M2
M1
M0
K7
The (B−V ) color of stars as a function of temper-
K5 ature (the points are coded by spectral type).
K4
This color index can be used to determine the
1.0

K3
K1 spectral type and hence effective temperature of
(B-V)

K0
G8 stars. The temperature scale runs from high
G2
G0
values (≈ 45,000 K) on the left to low values
0.5

F8
F5
(≈ 3000 K) on the right.
F2
F0
A8
A5
A2
0.0

A1
A0
B9
B7 B8
B5 B6
B3
B1 B2
B0
O5O6O7O8

4.50 4.25 4.00 3.75 3.50

log Teff

M7
M6
K7 M0M1 M5
M2
M3
M4

This figure has the same format as the previous

1

K5
K4
K3 one, except that the color index displayed is
(U − B). Unlike (B − V ), the (U − B) color index
K1
K0 for stars is multi-valued in temperature at some
G8 colors, and hence cannot always be used to deter-
(U-B)

A2
A5 A8 G2
F0F2 F8G0
mine a star’s temperature/spectral type. The
A1
0

A0 F5
large “twist” near 10,000 K (A stars) is due to the
B9
influence of Balmer absorption; the high-order
B8

B6
B7 Balmer lines overlap in the U band, significantly
B5
supressing the flux in this filter at these temper-
B3
B2
atures.
B1
-1

B0
O7O8
O5O6

4.50 4.25 4.00 3.75 3.50

log Teff

As one moves to temperatures larger than 10,000 K, the hydrogen becomes ionized, the Balmer lines weaken dramat-
ically, and the (U − B) color index resumes its expected march to the blue. (There is a small reversal of the (U − B)

691
curve at temperatures near 4000 K; this feature, however, is not due to blanketing from atomic hydrogen.) Like (B −V ),
(U − B) reaches a limit at extremely hot temperatures (i.e., when the peak of the spectral energy distribution lies at
wavelengths well shortward of the U band).

This series of optical stellar spectra displays the

systematic change in absorption line strengths
with temperature. The hottest stars (top
of diagram) possess virtually no indication of
hydrogen, because most of the atoms of that
element are ionized. As one drops down
the temperature sequence, the strength of the
hydrogen lines increases, reaching a maximum
at spectral type A0 (Hγ and Hβ are the lines
in the blue and green regions, respectively). At
cooler temperatues, the hydrogen lines weaken,
and various metal lines (e.g., Ca II, Na I) and
molecules (e.g., TiO) begin to dominate the spec-
trum.

One surprise (not shown in the spectra in previous viewgraphs): the spectra of the coolest stars (M stars) are dominated
by absorption due to molecular bands (TiO in particular)! The accompanying table lists the most important absorption
features as a function of spectral type.

Harvard Spectral Classification

Spectral
Type Primary Spectral Features

O Hottest stars (blue); few, weak lines

He II absorption; occasional emission lines
B Blue-white stars; strong Balmer absorption
He I absorption (strongest at B2)
A White stars; strongest Balmer lines at A0
Ca II lines appear
F Yellow-white stars; Ca II strengthens
Balmer lines weaken
G Yellow stars; solar-type spectra
Ca II, Fe I, and neutral metal lines strengthen
K Orange stars; Ca II strongest at K0
Spectra dominated by metal absorption lines
M Red stars; strong metal lines
Molecular bands dominate, in particular TiO

One final note before we leave this topic: always remember the importance of the value of kT ! This expression provides
the characteristic energy of particles in a system; this energy, in electron volts, is

T
E = kT = 8.6173 × 10−5 T eV = eV
11, 605

If you are interested in transitions that have energies of the order of an electron volt, to activate these transitions the
temperature must be an appreciable fraction (of order 10%) of 11,605 K; at room temperature (less than 3% of this
value), such transitions are strongly suppressed by the exponential term in the Boltzmann equation.

692
 Low-Mass Stars and Substellar Objects
Until 1995, few stars were known with spectral types later than M6. The coolest stars were classified as M9.5; their
spectra were characterised by quite strong TiO and VO absorption bands and little flux emerging at wavelengths below
7000 Å. The spectrum in the figure is that of an M9.5 star discovered in 1991 (this object also has strong Balmer
emission lines), and had an estimated effective temperature of less than 2000 K!
7000
6000

PC 0025+0447 A low-resolution spectrum of PC0025+0447, a

5000

late M-type star (M9.5) with strong Balmer emis-

sion discovered in 1991. This object may have
fnu 1 DN = AB 28

insufficient mass to initiate fusion in its core, and

4000

thus is not technically a “star”. PC0025+0447 is

thought to lie near the boundary between normal
3000

stars and brown dwarfs. The strong, wide absorp-

tion features are due to TiO and VO.
2000
1000
0

5000 6000 7000 8000 9000

Wavelength (A)

This paucity of cool star detections could be due to either a true cutoff in the mass distribution of stars (i.e., stars
with masses less than ≈ 0.1 M⊙ are exceedingly rare) or our inability to detect objects at this temperature. The latter
explanation was more popular, for if these cool objects did exist they would emit most of their radiation at wavelengths
beyond 1µ (i.e., in the infrared), and thus would be extremely difficult to detect with the instrumentation of the time.
If Nature did provide a continuous distribution of masses from the M-type stars all the way down to the planetary scale
(recall MJupiter = 0.001M⊙ ), at some mass limit the pressure and density in the object’s center would become too
low to ignite nuclear fusion. Such objects would no longer be “stars” in the standard sense; their luminosity would be
powered by gravitational contraction and the stored thermal energy from formation (which, of course, was also provided
by gravitational potential energy). In the 1960s these hypothetical objects were assigned the name of “black dwarfs”,
but this term is now used to describe another type of object (a cold white dwarf); the term “brown dwarf” (suggested
by Jill Tarter in 1974) has been chosen to describe this class of low mass, “failed stars”.
The first plausible candidate (found in 1988) for a brown dwarf was a companion to the star GD 165; the companion
had many of the expected traits of a brown dwarf, but the observations were not conclusive on this issue. In 1995
two definite brown dwarfs, Tiede 1 (in the Pleiades) and Gl 229B (the faint companion of the nearby star Gl 229)
were shown to possess all of the characteristics of a brown dwarf. In the last years of the 20th century large-area,
infrared-sensitive surveys (e.g., the Two-Micron All-Sky Survey [2MASS] and the Sloan Digital Sky Survey [SDSS])
began to identify large numbers of objects cooler than M-type stars (several hundred brown dwarfs have now been
found, many of which are “field” brown dwarfs, i.e., are not associated with other stars or clusters, but are isolated
objects in interstellar space). Because of the results of these investigations, the spectral sequence has been extended
beyond M to include two new classes: “L” and “T”.
In L dwarf spectra, the strong molecular metallic oxide bands (e.g., TiO) gradually weaken, such that by L8 they have
vanished; these features are replaced by absorption produced by metallic hydrides and neutral alkali elements. The
spectra of T dwarfs are dominated by strong methane (CH4 ) and water (steam!) bands; the existence of methane (one
of the main sources of absorption in Jovian planet atmospheres) requires that the atmospheric temperature of T dwarfs
be 1000 K or less! Spectra of stars from M6 to T8 are displayed in the accompanying figure.
The next step in the stellar sequence has already been assigned the name “Y”. These hypothetical objects, with
expected temperatures of hundreds of Kelvin, were eagerly searched for in the first decade of this century, and in 2011
announcements of the discovery of a number of Y dwarfs were made; the lowest temperture Y dwarf, identified by a
team led by Penn State’s Kevin Luhman, is, in the words of the press release, “Cooler than Earth”! In 2013 Luhman

693
 M6 L8

M7 L9
20

M8 T0

M9 T1

L0 T2
15

L1 T3
Normalized fλ

L2 T4
10

L3 T5

L4 T6

L5 T7
5

L6 T8

L7
0

10000 15000 20000 25000 10000 15000 20000 25000

Wavelength (A) Wavelength (A)

Spectra of late-type stars (spectral types M6 to T8). The horizontal lines associated with each spectrum are the zero
flux density levels. The L and T objects cannot produce energy via hydrogen fusion, and are often referred to as “brown
dwarfs” rather than “stars”. The “noisy” regions in a few of the spectra near 14,000 Å and 18,500 Å are not features
in the objects but low S/N data produced by strong molecular absorption bands in the Earth’s atmosphere.

discovered a binary brown dwarf system at a distance of only 2 pc (now designated as Luhman-16); this is the third
closest known stellar system, and the nearest stellar system found in nearly a century!
The defining characteristic of brown dwarfs is their mass; the lowest brown dwarf mass is 13 MJ (objects with less mass
cannot ignite deuterium fusion) and the maximum mass is about 80 MJ (where it becomes possible to fuse hydrogen).
Objects with M < 13MJ are classified as planets (down to some undefined mass limit). All brown dwarfs have similar
radii to that of Jupiter (indeed, Jupiter’s radius is near the maximum of the allowed radius for a brown dwarf). A
brown dwarf, unlike a star, does not have a stable appearance. The sun’s temperature and luminosity are relatively

694
 A map displaying the locations of the four nearest
star systems to the sun. The distance to the
closest system, α Centauri, was determined in
1838, and the red dwarf Proxima Centauri, which
is slightly closer than the two brighter main
sequence stars, was first identified by Robert
Innes in 1915, and its parallax measured two
years later by Joan Voute. In 2013 and 2014
the distances to two brown dwarfs were found
by Kevin Luhman to lie at distances just beyond
Barnard’s star, whose parallax was measured
in 1916.

stable for billions of years, and that of M stars for the age of the universe, but a brown dwarf’s temperature (and hence
luminosity given the constant radius) changes considerably over its lifetime.
Because a brown dwarf’s source of luminosity is gravitational contraction and not nuclear fusion, there is no mass-
luminosity relation for these failed stars; brown dwarf luminosities are primarily a function of time (they gradually
cool and thus fade). Brown dwarf luminosities are quite low; a 1500 K, Jupiter-sized object is 20,000 times fainter
(10.75 magnitudes) fainter than the sun; the most extreme cases, where the effective temperatures are 300 K, the
luminosities are down by a factor of over ten million on that of the sun!
The study of brown dwarfs is an extremely exciting field; the continuing advance of infrared instrumentation promises
dramatic increases in our knowlege of substellar objects in the near future. We are only just beginning to probe the
mysteries of these exotic objects that exist in the realm between stars and planets.

The H–R Diagram

Early in the 20th century, astronomers had determined the properties of a reasonably large sample of stars. It was
clear that on average the O stars were hotter and more luminous than M stars. The first attempts at measuring
the mass-luminosity relation had been completed; combining this information led to the (generally correct) belief that
O stars were the most massive, M stars the least massive, and that O stars had larger radii than M stars.
Thus the spectral classification scheme was interpreted as an age sequence. Remember that at this time the energy
source for the stars was unknown. A popular theory held that stars started their lives as hot, massive O stars. As stars
burned their fuel (whatever that was), they would lose mass, drop in luminosity and temperature, and end their lives
as M stars (or, more probably, continue to shrink and cool and slowly disappear from view). This scenario is the origin
of the terminology early-type and late-type stars. In this view the sun is currently somewhere near the middle of its
lifespan.
The Danish chemical engineer and amateur astronomer, Ejnar Hertzsprung, analyzed the properties of a few dozen stars
with reasonably accurate parallaxes, and in 1905 published a classic paper (“The Radiation of the Stars”) demonstrating
the strong correlation between spectral type (which corresponds to temperature) and luminosity (which ties to the
mass of the star via the mass-luminosity relation). The tightness of the observed correlation between luminosity and
temperature implied that the radius of a star was not a free parameter but also depended upon the star’s temperature.
If the mass was related to the luminosity, this relationship indicated that there was something fundamental about how
the mass of the star influenced the star’s structure; taken to the extreme, one could say that given the mass, everything
about the star was known!
While this model was an attractive picture, the data revealed one anomaly. In 1907 Hertzsprung reported that although
the majority of stars did follow the relationship (hot stars were more luminous), for stars of G-type (similar to the sun)
and later there were a few objects that were significantly more luminous (up to a factor of ≈ 100!) than expected given
their temperature.
Since the stars of a given spectral type have the same temperature, it is obvious that the more luminous stars must
have larger radii:
4πR22 σT24
r
L2 R2 L2
= ⇒ =
L1 4πR21 σT14 R1 L1

Since there is a range of more than a factor of 100 in the luminosity at some spectral types, some stars must be more
than ten times larger in linear scale than the sun! Hertzsprung designated the “overluminous” objects “giant stars”.

695
In 1911 Hertzsprung extended his work by comparing the magnitudes and colors of stars in clusters, and found the
same phenomenon. This approach, measuring the color-magnitude relation for clusters, is a powerful technique, and
is an active field of research today. The basic idea is that all of the stars in the cluster have the same distance from
Earth (the physical size of the cluster is insignificant compared to the distance between the Earth and the cluster),
hence all have the same distance modulus, so when comparing the cluster stars with each other, measuring the apparent
magnitude (easy) works as well as the absolute magnitude (hard to determine) for internal comparisons. We will discuss
this subject in detail later in the lecture; this technique is but another example of Hertzsprung’s genius!
This work was rapidly recognized by European scientists as of fundamental importance, and Hertzsprung, despite no
formal training in astronomy, became a world leader in the field, eventually rising to become the Director of one of
the top astronomical centers, the Leiden Observatory. Hertzsprung also made one of the first effective efforts (in 1913)
at determining the distances to what we now call galaxies at a time when the Milky Way was considered to be ”the
Universe” (to be discussed later in the course). We professional astronomers should keep in mind that a number of the
great astronomical discoveries of the last century were made by an engineer!

Ejnar Hertzsprung (1873-1967) and Henry Norris

Russell (1877-1957) independently discovered the
correlation between stellar luminosity and spec-
tral class (temperature); the graph displaying
this relation is known as the Hertzsprung-Russell
Diagram, or more frequently as simply the “H-R
Diagram”. This figure became one of the most
important astrophysical tools to investigate the
characteristics and evolutionary history of stars.
Both of these scientists made a number of impor-
tant contributions to astronomy.

Henry Norris Russell of Princeton University, arguably the greatest American astronomer in history, was intrigued by
the same problem as Hertzsprung. (The highest prize of the American Astronomical Society, awarded for recognition
of lifetime contributions to astronomy, is the Russell Lectureship.) Ninety years ago there was no World Wide Web to
instantaneously circulate results, and Russell labored on the problem until 1913 in ignorance of Hertzsprung’s work.
(Hertzsprung’s discoveries had been published in the German journal Zeitschrift für Wissenchaftlichte Photographie
[Magazine for Scientific Photography].)
Russell reached the same conclusions as Hertzsprung, but Russell immediately created a figure showing absolute magni-
tudes against spectral type for more than 200 stars in the solar neighborhood. His graph revealed that the vast majority
of stars lay in a band running from hot, luminous stars to cool, faint stars, with about 10% of the objects of relatively
late spectral types that were “overluminous” by a factor of about a hundred. Like Hertzsprung, Russell designated the
unusually luminous stars giants; Russell called the “normal” stars dwarfs.
It is known that Hertzsprung did construct at least one graph of stellar luminosity as a function of temperature, but,
difficult as it is to believe, this type of figure was not included in any of his early publications; his papers contained
many tables with a multitude of columns. If Hertzsprung had simply added a figure to one of his seminal works that
graphically displayed luminosity as a function of spectral type, it is quite likely that today we would be studying
Hertzsprung, not Hertzsprung-Russell, Diagrams! The first “Russell” Diagram contains most of the main features of
its modern successor, the Hertzsprung-Russell Diagram (see figure).
The faint star in the lower left of Russell’s figure (10 magnitudes, or a factor of 10,000 in luminosity, fainter than
expected for a star of this temperature) has a size much more typical of the Earth than the sun! Russell thought that
this faint, hot star was a rare case of a subluminous object; in fact, this star (40 Eriandi B) was one of the first known
white dwarfs, and in the coming decades this class of objects would become the center of one of the great scientific
controversies of the last century.
This type of graphical representation of the properties of stars has widespread use in astronomy today. The expression
“H-R Diagram” technically means a plot of absolute V magnitude as a function of spectral type, but the term is applied
to a wide variety of similar displays. Another common name for this presentation is “Color-Magnitude Diagram” (CMD).

696
 Henry Norris Russell’s first construction of what
would become known as the Hertzsprung-Russell
Diagram, a key diagnostic for stellar evolution.
Russell’s work, although completed several years
later than that of Hertzsprung, was independent
of that of the Danish engineer. The basic prop-
erty of the H-R Diagram is the relation between
the luminosity of a star and its surface tempera-
ture (these quantities are represented by absolute
visual magnitude and spectral class in Russell’s
original work). The majority of the stars appear
to lie in a relatively confined diagonal line from
high luminosity, high temperature to low lumi-
nosity, low temperature. There are about a dozen
stars, later than type G, that are significantly
brighter than their spectral class counterparts on
the main sequence; the straightforward interpre-
tation is that they are much larger than main
sequence stars. There is one A-class star that lies
about ten magnitudes below the main sequence
A stars, suggesting that its radius is 100 times
smaller than its main sequence counterpart.

Features of the Original Russell Diagram

• Axes: Spectral Type and Absolute V Magnitude.

• Most stars (80–90%) lie in a band running from the upper left (hot, luminous O stars) to the
lower right (cool, faint M stars). This feature is called the main sequence; stars in this region are
known as dwarfs. The sun is a G2 dwarf.
• A group of giants, mainly cooler than F stars, populates the upper right of the diagram; giants
have absolute magnitudes around 0 (nearly 100 times more luminous than the sun!)
• In the lower left there is an MV = +11 A-type star, which is over 10 magnitudes fainter than a
main sequence A star; this factor in 10,000 in luminosity translates into a factor of 100 in radius.

• H-R Diagram is now often called a “Color-Magnitude Diagram”

• Abscissa: Spectral Type, Color Index, Effective Temperature
L
• Ordinate: MV , Mbol , L⊙
, V, B

An example of commonly used variables for H-R/Color-Magnitude Diagrams

Abscissa: Spectral Type, Color Index (but not (U − B)!), Effective Temperature
Ordinate: MV , Mbol , LL , V , B
⊙
The four panels in the accompanying figure present some of the basic properties of main sequence stars as a function of
effective temperature (which is, as can be seen from the coding of the points, closely related to the spectral type). The
upper left figure shows the absolute bolometric, not absolute V , magnitude; in this diagram one can see the enormous
luminosities (hundreds of thousands of solar luminosities) of the O stars! The upper right panel displays the strong
dependence of the bolometric correction on temperature (a feature we suggested existed in our previous discussion

697
 -5
O5O6O7
O8

B0
B1
B2 B3 A modern version of the Hertzsprung-Russell
B5
B6 Diagram for the main sequence, with the abso-
0

B7
B8 B9
A0A1
A2
lute V magnitude plotted against the effective
A5
A8
temperature. The points are coded by the
F0
F2
F5 stellar type. The apparent maximum luminosity
F8
G0
of O stars and the precipitous luminosity drop
MV

G2
5

G8
K0
K1
K3
K4
in M stars are artifacts of observing in a fixed
K5
K7 bandpass (the V filter), the O stars and M stars
M0 have much larger bolometric luminosities (rela-
M1
10

M2 tive to the sun) than suggested in this figure.

M3
M4
M5
M6
M7
15

4.50 4.25 4.00 3.75 3.50

log Teff

of the bolometric correction); for A0 through K1 stars the adjustment from V to bolometric magnitudes is relatively
modest (much less than a factor of two in luminosity), but at both extremes the corrections exceed a luminosity factor
of twenty!
The mass and radius relationships, presented in the lower two panels, confirm that the hottest stars are more massive
(approaching 100 M⊙ ) and larger (over 10 R⊙ ) than the sun, while the coolest M stars are considerably less massive
and smaller the the sun (M7 stars have masses of approximately 0.15 M⊙ and radii of ≈ 0.2 R⊙ ). As previously
mentioned, the correlation of a main sequence star’s luminosity, temperature, and mass requires that the mass and
radius be related; an approximate expression is R ∝ M 0.6 .

In principle it is difficult to construct an H-R Diagram, as one usually does so one star at a time (measure an apparent
magnitude in two filters, determine a parallax). Hertzsprung had a brilliant insight to circumvent this observational
complication: examine a cluster of stars. If one obtains magnitudes in two filters (typically B and V ) and plot one of the
magnitudes against the color index, one can produce an H-R Diagram of the cluster without any parallax measurements!

The color-magnitude diagram for the globular

cluster M15 (a collection of stars located in our
galaxy) constructed from Hubble Space Telescope
B and V observations. Since the stars are essen-
tially all at the same distance (the radius of the
cluster is a tiny fraction of the cluster’s distance
from Earth), the V magnitude is simply MV plus
a constant (5 log d − 5). The main sequence is the
diagonal line of stars starting near (B−V ) = +0.5
and sloping to the lower right corner of the panel.
The stars that lie above that point (with a wide
range of colors) are objects that have evolved
off of the main sequence; we will discuss these
objects in about a month. The sun’s (B − V )
is +0.65, so its counterpart in M15 has V ≈ 22.
What is the distance to the cluster?

This technique has a tremendous advantage over the one-star-at-a-time approach. One can measure magnitudes and
colors of hundreds of stars with just two observations (plates taken through two filters)! Since all the stars are at
(essentially) the same distance, there is a one-to-one correspondence between apparent and absolute magnitudes (a
single constant [5 log d − 5] is all that is required to transform apparent to absolute magnitudes in the band for all

698
 O5
O6

0
O7 F0
A8F2 G0
O8 A2A5 F5F8
A1 G2G8
A0 K0
K1
B0 B9 K3
K4
K5
B1 B8
-5

-1
B7 K7
B2
B3 B6
M0
B5

Bolometric Correction
B5 M1
B6
B7 M2
B8 B3

-2
0

B9 M3
Mbol

A0A1
A2 B2 M4
A5
A8
F0
F2 B1 M5
F5
F8

-3
G0
G2
B0
5

G8
K0 M6
K1
K3
K4
K5 M7
K7 O8
M0 O7
M1
M2 O6

-4
M3
M4
10

M5
M6 O5
M7
4.50 4.25 4.00 3.75 3.50 4.50 4.25 4.00 3.75 3.50
log Teff log Teff

O5
O6
O5 O7
1.0 O8
O6
O7 B0
O8
B0 B1
B1 B2
B3
1

0.5

B2 B5
B3 B6
B7
B5 B8B9
B6 A0
log M/Msun

log R/Rsun

B7 A1
A2
B8 A5
B9
A0 A8
A1 F0
A2 F2
A5 F5
A8 F8
0.0

F0 G0
G2
F2F5 G8
K0
K1
K3
F8 K4
K5
G0
G2 K7
0

G8
K0 M0
K1
K3
K4
K5 M1
K7
M0 M2
M1 M3
M2 M4
-0.5

M3
M4 M5
M5
M6 M6
M7
M7

4.50 4.25 4.00 3.75 3.50 4.50 4.25 4.00 3.75 3.50
log Teff log Teff

Four properties of main sequence stars (absolute bolometric magnitude, bolometric correction, mass, and radius) as a
function of effetive temperature. The points are coded by spectral class; the sun is represented by spectal class G2.

stars in the cluster). Thus one does not need to know the distance to make this graph, so no parallax measurements
(extremely difficult) are required. One does not need to make any bolometric corrections for this technique to work;
if one obtains the cluster data in the same filter that was used in the main sequence calibration work, the magnitude
offset between the cluster and standard main sequences will be the distance modulus. The only assumption is that
the cluster and standard main sequences are the same; while this is largely true, there are subtle differences (which we
won’t worry about in this class).
This technique is so well calibrated that astronomers use the vertical (brightness/luminosity) offset between the observed

699
cluster HRD and the “true” one (the difference between measured apparent and the absolute magnitudes based upon
calibrations of nearby stars) to measure the cluster’s distance modulus and hence to determine the distance to the
cluster! For example, one assumes that a cluster star on the main sequence that has the same (B − V ) as the sun is
identical to the sun in every way. Since you can infer the absolute magnitude of the star from the color, and know the
apparent magnitude from observation, the distance can be easily calculated from the distance modulus.

Summary of the Cluster HRD Distance Method

1. Observe cluster in two filters that form a useful color index for conversion to spectral type.
2. Construct HRD of apparent magnitude vs color index for stars in cluster.
3. For each star on the main sequence, one knows its absolute magnitude from locally-based HRDs.
4. The distance modulus to the cluster (5 log d − 5) is simply the difference between the observed
magnitude and the absolute magnitude of the star.

For example, if you formed a CMD of a cluster, and found that a star on the main sequence with the same (B − V ) as
the sun had V = 14.80, you would immediately know that the distance modulus of the cluster was 10.0, or the distance
was 1000 pc!
Parallaxes in 1900 were only useful for distance measurements out to 50–100 pc; today with digital detectors and clever
image processing software one can push this limit out over 1000 pc. With the cluster technique, distances out to 10 kpc
were routinely done by the middle of of the last century. Today, assisted by observations by space missions, cluster
HRDs can be used to measure cluster-based distances up to several Mpc!

Luminosity Classes
In 1897 Antonic Maury noticed that stars of a given spectral type could have distinctly different spectra: in particular the
widths of the absorption features varied. Hertzsprung speculated that this variation might make it possible to separate
giants from dwarfs based solely on their spectra. He suggested, based on an early analysis, that the narrow-lined stars
were more luminous than the broad-lined ones.
This proposal sounds like a wild idea, but time proved Hertzsprung correct. (Why would increasing the luminosity of
a star decrease the width of the line?) This discovery is a good model for those of you who wish to have a career in
astronomy. When you have an idea, publish it. Most of the time you will be wrong, but the furor will die in six months;
but if you are right but once, it will be remembered forever and your brillance will be celebrated by generations of
future students!!
For much of the first half of the 20th century, astronomers made careful comparisons of stellar spectra, and found
that there were indeed subtle differences that clearly correlated with stellar luminosity. In 1943 William W. Morgan,
Philip C. Keenan, and Edith Kellman of Yerkes Observatory published their Atlas of Stellar Spectra, with an Outline
of Spectral Classification, adding another dimension, luminosity class, to the description of a star. The atlas contains
55 spectra that clearly display the effects of temperature and luminosity in the spectra of stars. Their characterization
of the properties of stars was designated the “MKK system”.
Ten years later Morgan and Keenan slightly revised stellar classification scheme; Their system (now known as the MK
Classification) is in use today (and will probably be employed for centuries). Stars are classified by two parameters:
their temperature (essentially the Cannon designation) and their luminosity. Stars on the main sequence (dwarfs) are
denoted by luminosity class V; the sun is a G2 V star while Vega is given the classification of A0 V. The more luminous
a star (at a given spectral type), the “lower” the Roman numeral associated with the luminosity class; a typical “giant”
star has a luminosity class of III while typical “supergiants” are luminosity class I. The most luminous designation has
undergone subdivision as techniques were developed to distinguish various luminosity levels among supergiants (alas,
as you know, Roman numbers have no zero!). Aldebaran, Arcturus, Pollex, and Capella are all luminosity class III
objects; Antares, Rigel, Deneb, and Betelgeuse are supergiants.
The accompanying figure is the HR Diagram [MV vs. (B − V )] for the stars in the Yale Bright Star Catalog that have
V < 5.0, have measured parallaxes, and are bluer than (B − V ) < +2.2 (this last restriction is to avoid scaling issues
due to the presence of a few extremely red stars). The brightest stars are labelled in red; the sun is in blue. These

700
 A photograph of William W. Morgan (1906-1994)
who, with Philip C. Keenan (1908-2000), both of
Yerkes Observatory, developed the “MK Classi-
fication System” that is now widely used today.
This scheme extends Cannon’s spectral type into
the dimension of luminosity, with main sequence
stars assigned luminosity class V and supergiants
luminosity class I. The MK classification of the
sun is G2 V.

M-K Luminosity Classes

Type Description Type Description

Ia-O Extreme, luminous supergiant IV Subgiants

Ia Luminous supergiant V Dwarfs (main sequence)
Ib Less luminous supergiant VI, sd Subdwarfs
II Bright Giants D White dwarfs
III Giants

A3 I
0

The spectra of three stars, all of spectral type A3,

but of different luminosity classes. The topmost
spectrum is a supergiant (I), the middle spectrum
1
Normalized AB magnitude

A3 III
is a giant (III), and the bottom spectrum is a
main sequence, or dwarf, star (V). The difference
between the supergiant and the other two stars
2

is obvious (the supergiant lines are in general

A3 V
weaker and narrower than those in a dwarf); the
variations between giants and dwarfs are more
3

subtle.
4

4000 4250 4500 4750 5000

Wavelength (A)

objects are divided into two groups: the main sequence (diagonal line on the left) and the giant branch (broad clump
on the right). From this diagram one would infer than roughly half of the stars are giants and the sun is one of the
least luminous stars, whereas in fact the vast majority of stars are dwarfs, and are fainter than the sun! This skewed
view of stellar populations is an observational effect: to appear on this diagram, a star similar to the sun must have a
distance no greater than about 10 pc, whereas luminous supergiants such as Deneb easily satisfy our flux limit at 5 kpc!
(One way to describe this situation is that our eyes scan a volume over 100 million times larger for stars like Deneb
than stars such as the sun; if the night sky contained one star like the sun and one star like Deneb, the true number
density of stars similar to the sun would be 100 million times larger than the number density of Deneb-type stars, not
the one-to-one ratio inferred from observation.) We will discuss these “selection effects” later in the course, as they

701
 This figure
Deneb
presents the Hertzsprung-Russell
Diagram for the brightest stars in
the sky. The individual points
are objects from the Yale Bright
Betelgeuse
Star Catalog that have V < 5.0,
have measured parallaxes, and
-5

B Centauri
Rigel
are bluer than (B − V ) < +2.2.
Acrux
Polaris The main sequence is the diag-
Canopus
onal line on the left; the giant
Absolute V Magnitude

Achernar branch is the loose clump on the

Spica Antares
right.
The sun’s location is indicated in
Aldebaran
Regulus
Capella
Arcturus
blue (near the bottom of the main
sequence, just below α Centauri);
0

Vega
Pollux
virtually every star that you can
Castor
Sirius see in the night sky is more lumi-
Fomalhaut
Altair
nous than the sun! From this
Procyon diagram you might conclude that
stars are approximately evenly
divided between the giant and the
a Centauri
Sun dwarf classes, whereas in fact the
5

dwarfs are much more numerous

Yale Catalog V < 5.00 than giants. (The distance limit
that we can see dwarfs is much
smaller than that of giants.)

0.0 0.5 1.0 1.5 2.0

(B - V)

frequently arise whenever one deals with flux-limited samples.

Again, one needs to take care in interpreting MV as a luminosity. For stars such as the sun, MV ≈ Mbol , but for hot
objects (e.g., Rigel) or cool ones (e.g., Betelgeuse), the bolometric correction can be several magnitudes!
The accompanying figure displays the spatial distribution stars from the Yale Bright Star Catalog that have V < 5.0
and have measured parallaxes. (This is actually a two-dimensional representation; a star’s distance from the center of
the figure is related to the three-dimensional distance from the sun.) The sun is the blue object in the center, the two
blue circles are distances of 10 pc and 100 pc, and the outer circle is a distance of 1 kpc. The angular distribution is
set by the right ascension of the star (the RAs are the numbers just inside of the outer circle). In order to produce a
visually useful distribution, the distance scale is a combination of linear and logarithmic scales: the distances to stars
located within 10 pc of the sun are on a linear scale; those outside of the 10 pc circle are placed on a logarithmic scale.
The α Centauri system is the star nearest the sun (located at near right ascension 14h ; the brightest star in this multiple
system has a classification of G2 V). Sirius (A1 V, the second closest star in this figure) and Procyon (F5 IV-V) are
the two stars near α = 7h , and Deneb, a first-magnitude A2 Ia star nearly a kpc from the sun (!), is at α = 20h 47m .
From the diagram you can see that the typical distances to the visible stars are on the order of 30 pc; given that the
apparent magnitudes of these stars are V ≈ 3, the typical luminosities are MV ≈ +0.6, much more luminous than the
sun. (Indeed, if the sun was located more than 11 pc from the Earth, it would not appear in this diagram!)
The next figure attempts to provide a visual impression of the enormous range of scales in the sizes of familiar celestial
objects via a “Powers of Ten”-type display in six panels. In each panel the leftmost object is the rightmost object in
the previous display. One begins in the upper left with the terrestial planets, moving to the Jovian Planets, dwarf stars,
giant stars, “small” supergiants, and “large” supergiants. The radii of the objects vary from 0.384 R⊕ to 6.6 AU, a
factor of approximately 400,000 in linear scale (a factor of 6.6 × 1017 in volume)! The supergiant UY Sct, with a radius
of 7.94 AU, is the largest known star.
The M-K Classification scheme allows one to determine an (approximate) absolute magnitude without measuring
a parallax. If one measures the apparent magnitude (easy) and the M-K Class from a spectrum (harder, but not

702
 06
07 05

08 04

09 03

The spatial distribution of stars

from the Yale Bright Star Catalog
10 02
that have V < 5.0 and have
measured parallaxes. The sun
is the blue object in the center,
11 01 the two larger blue circles are
distances of 10 pc and 100 pc,
and the outer circle is a distance
12 00 of 1 kpc. The angular distribu-
tion is set by the right ascen-
sion of the star (the RAs are
the numbers just inside of the
13 23
outer circle). The distance scale
is complex and highly non-linear;
see text for a description.
14 22

15 21

16 20

17 19
18
Yale Catalog V < 5.00

A series of six panels displaying the enor-

mous range of sizes of objects from planets
to supergiant stars (similar to the familiar
”Powers of Ten” movie). 1) Terrestrial Planets:
Mercury, Mars, Venus, Earth; 2) Jovian Planets:
Earth, Neptune, Uranus, Saturn, Jupiter;
3) Dwarf Stars: Jupiter, Wolf 359, Sun, Sirius;
4) Giant Stars: Sirius, Pollux, Arcturus, Alde-
bran; 5) “Small” Supergiants: Aldebran, Rigel,
Antares, Betelgeuse; and 6) “Large” Supergiants:
Betelgeuse, µ Cephei, VV Cephei A, VY CMa.
The last-named star has a radius of 6.6 AU!

extremely difficult), one has both the apparent and absolute magnitude, so the distance modulus can be calculated.
This procedure was a tremendous advance in the field (distances to individual stars without a parallax measurement,
as well as dramatically extending the distance scale). Most stellar distances in the 20th century were determined by
this technique, which is called the spectroscopic parallax.
The major drawback of this method is its limited accuracy. One can see from the accompanying diagrams that the
scatter in absolute magnitude for stars above the main sequence (luminosity classes I–IV) is about one magnitude per
luminosity class (i.e., one can determine the absolute magnitude of a given non-main sequence star to only about one
magnitude). This luminosity error directly translates into a distance error. An offset of a magnitude produces a change

703
 √
of a factor of 2.512 in luminosity; this result translates into an error of about a factor of 1.6 ( 2.512) in the distance.
This technique does not provide a particularly accurate measurement of a stellar distance, but, as Mark Twain said
about old age, “It is indeed terrible, unless one considers the alternative.”
Why are the features in a given spectral class different in giants and dwarfs? One would expect the spectra to be
identical for a given spectral type because the temperatures should be the same. To investigate this issue, let us
examine a slightly different H-R Diagram. Theorists prefer to plot the logarithm of the bolometric luminosity as a
function of the logarithm of the effective temperature (physical properties compared to observational quantities). The
points presented in this fashion have the same shape as the standard H-R diagram; after all the magnitude is simply the
logarithm of the luminosity, and the logarithm of the temperature roughly corresponds to the spectral type sequence.
This diagram immediately demonstrates that hot stars on the main sequence are larger than cool main sequence stars,
but the range in radii on the main sequence is relatively small compared to the range in luminosities. Hot main sequence
stars are many times the sun’s diameter, while cool main sequence stars are several times smaller than the sun. The
giants, however, are much larger than main sequence stars; a typical giant has a radius of around 10-30 R⊙ .

O5 R=
O6
O7 100
O8 Ma R
in sun
Supergiants
Se
B0 qu
en
B1 ce
4

B2
The Hertzsprung-Russell Diagram presented as
B3 the luminosity vs. effective temperature. The
R=
B5
B6 Giants three diagonal lines represent constant radii; the
R B7
sun lower one is for objects with a size comparable
2

B8
log L/Lsun

B9
A0
A1A2
to the Earth, the middle one is where stars with
A5
A8 sizes identical to the sun would lie, and the upper
F0
F2
F5
F8
one is for objects with radii 100 times larger than
G0
G2
the sun (0.465 AU).
0

G8
K0
K1
K3
K4
K5
K7
M0
R= M1
M2
0.0 M3
1R White Dwarfs M4
-2

sun M5
M6
M7

4.50 4.25 4.00 3.75 3.50

log Teff

Even more amazing than the giants are the supergiants; their luminosities range up to MV ≈ −9! For G supergiants,
this luminosity implies a radius of over 500 solar radii!! Betelgeuse, an M supergiant, is estimated to have a radius
of ≈ 900 R⊙ (about 4 AU!); the radius of the star UY Sct is believed to be ≈ 7.9 AU! (If UY Sct was placed in the
center of the solar system, its surface would be located between the orbits of Jupiter and Saturn!!)
What are the average densities and surface gravities of supergiants? Let us assume that a supergiant radius is 300
times that of the sun. Then
   R 3  R 2
ρ M ⊙ M g M ⊙ M
= = 4 × 10−8 and = = 1 × 10−5
ρ⊙ M⊙ R M⊙ g⊙ M⊙ R M⊙

It is clear that unless supergiants have truly extraordinary masses (on the order of a million times the mass of the sun!),
they will have much lower densities and surface gravities than that of the sun. (The same is true of giants, but the
ratios are on the order of a factor of 1000 less extreme.) There are no known stars with masses larger than ≈ 100 M⊙
(more on this issue in a few weeks); the mass of a typical supergiant is thought to be around 10–20 M⊙ .
These calculations reveal that the densities and surface gravities of supergiants are much less than those for main
sequence stars. This result has two consequences for spectral line formation, neither of which we will discuss in detail
in this course.
The main luminosity dependence in the spectrum is the decreasing width of the lines as the luminosity rises. This
change is produced by an effect called pressure broadening that is well understood (high pressures alter the energy
states of the atomic transitions), but a discussion of this topic is beyond the scope of this course. A brief summary:
in our previous discussions of energy levels in atoms, we implied that each level has a unique value of the energy. If
this is true, then all bound-bound transitions can be associated with a single energy. Quantum mechanics, however,

704
teaches us that there is a fundmental uncertainty associated with the energy of a particle (one manifiestation of the
famous Uncertainty Principle), therefore each energy level has a finite (usually small) width of permitted energies, thus
a specific transition will have a finite range of energies. The energy widths of the levels widen with increasing pressure,
thus the photons produced (or absorbed) by a given transition will have a range of wavelengths that increases with
increasing pressure, broadening the observed line profile.
The lower densities and gravities in the atmospheres of giant stars produce lower pressures in the region of line formation.
Recall that we mentioned that the ionization state of an element depends upon the pressure due to free electrions (more
pressure, less ionization). Thus at a given temperature the atoms in a supergiant atmosphere will be more ionized than
those in a main sequence star; this effect introduces subtle shifts in the relative strengths of lines.
The main point to remember: Increasing the pressure increases the line width and decreases the level of ionization,
therefore main sequence stars, with higher atmospheric pressures than giants and supergiants, will have broader lines
and lower ionization states than giants/supergiants at a given effective temperature.

705