Determination of Planck’s Constant Using the Photoelectric Effect

Lulu Liu (Partner: Pablo Solis)∗

MIT Undergraduate
(Dated: September 25, 2007)
Together with my partner, Pablo Solis, we demonstrate the particle-like nature of light as charac-
terized by photons with quantized and distinct energies each dependent only on its frequency (ν) and
scaled by Planck’s constant (h). Using data obtained through the bombardment of an alkali metal
surface (in our case, potassium) by light of varying frequencies, we calculate Planck’s constant, h
to be 1.92 × 10−15 ± 1.08 × 10−15 eV · s.

1. THEORY AND MOTIVATION This maximum kinetic energy can be determined by

applying a retarding potential (Vr ) across a vacuum gap
It was discovered by Heinrich Hertz that light incident in a circuit with an amp meter. On one side of this gap
upon a matter target caused the emission of electrons is the photoelectron emitter, a metal with work function
from the target. The effect was termed the Hertz Effect W0 . We let light of different frequencies strike this emit-
(and later the Photoelectric Effect) and the electrons re- ter. When eVr = Kmax we will cease to see any current
ferred to as photoelectrons. It was understood that the through the circuit. By finding this voltage we calculate
electrons were able to absorb the energy of the incident the maximum kinetic energy of the electrons emitted as a
light and escape from the coulomb potential that bound function of radiation frequency. This relationship (with
it to the nucleus. some variation due to error) should model Equation 2
According to classical wave theory, the energy of a above.
light wave is proportional to the intensity of the light
beam only. Therefore, varying the frequency of the light
should have no effect on the number and energy of resul- 3. EXPERIMENTAL SETUP
tant photoelectrons. We hope to disprove this classical
hypothesis through experimentation, by demonstrating 3.1. Schematic
that the energy of light does indeed depend on the fre-
quency of light, and that this dependence is linear with
Planck’s constant h as the constant of proportionality.

2. HYPOTHESIS

Light comes in discrete packets, called photons, each

with an energy proportional to its frequency.
E = hν (1)
For each metal, there exists a minimum binding energy
for an electron characteristic of the element, also called
the work function (W0 ). When a photon strikes a bound
electron, it transfers its energy to the electron. If this
energy is less than the metal’s work function, the photon
is re-emitted and no electrons are liberated. If this energy FIG. 1: Schematic of experimental setup with retarding volt-
is greater than an electron’s binding energy, the electron age applied. Photons are incident on the photocathode and
escapes from the metal with a kinetic energy equal to travel toward the anode to complete the circuit unless stopped
the difference between the photon’s original energy and by a high enough applied voltage.
the electron’s binding energy (by conservation of energy).
Therefore, the maximum kinetic energy of any liberated
electron is equal to the energy of the photon less the
minimum binding energy (the work function). Expressed 3.2. Apparatus
concisely the relationship is as such:
Our monochromatic light source was an Oriel 65130
Kmax = hν − W0 (2)
Mercury Lamp in combination with a narrow band pass
filter wheel with four different wavelength passbands.
The wavelengths are listed in Table I.
∗ Electronic address: [email protected]
 2

Wavelengths (nm) light. Its effects become most prominent when the re-
365.0 ± 2.0 tarding voltage is high. The retarding voltage is seen as
404.7 ± 2.0 an accelerating voltage by these electrons and they travel
546.1 ± 2.0
uninhibited from the anode to the cathode, opposite the
direction of the expected flow.
577.0 ± 2.0
Another feature that should not be neglected is the
TABLE I: Spectrum from Oriel mercury lamp. non-linear nature of the photocurrent vs. voltage curve
near the stopping voltage. Theoretically, at the stopping
voltage, we expect the current to be zero. However, as
A Leybold photocell served as our target, containing a we decrease the stopping voltage, the current rises only
potassium (W0 = 2.3eV ) photosurface as the cathode very slowly until it begins to take on a familiar I = αVr
and a platinum ring (Wa = 5.7eV ) as the anode sepa- linear form. This is to be expected since the number
rated by a vacuum. It was enclosed in a black box with of states with energy near the minimum binding energy
a small circular opening to allow for incoming light. Pre- N (E ≈ −W0 ) may in fact be very small, and increases
cautions were taken to shield the setup from ambient as the binding energy increases.
light, to protect the filters and photocell from overheat- Complicating effects such as the two represented above
ing, and to minimize illumination of the anode. compromise the reliability of zero-current crossings for
determination of stopping voltages. Instead, we look to
two different methods for extrapolating the data points
4. DATA AND ANALYSIS of interest. Any differences in results will be used in
calculation of a lower bound on our systematic error.
An example of our tabulated raw data is shown in Ta-
ble II for the 365.0 nm wavelength over five trials.
The normalized currents for each wavelength are plot- 4.1. Method One for Voltage Cut-Off
ted against their respective retarding voltages in Figure Determination: Linear Fit Method
2 with the standard deviations as error bars. The nor-
malization removes the scaling effects of the non-uniform
distribution of intensities across the spectrum of our light Demonstration of the Linear Fit Method of Stopping Voltage Approximation
40
source. With intensity normalized away, it is already ev-
ident from this figure that the cut-off voltages have some 35
Measured Values (I vs. V) for 404.7nm Photons
Low Voltage Best Fit Line
r

dependence on frequency. High Voltage Best Fit Line

Extrapolated Zero−Point
30
y = mx + b
m = −60.0 +− 5.1
b = 35.6 +− 1.5
25
Photocurrent Detected (pA)

2
Plot of Photocurrents Induced by Four Wavelengths of Light (Normalized to 577.0 nm Curve) X = 1.9
40
365.0 nm
404.7 nm 20
35 546.1 nm
577.0 nm
15
30

10
25 404.7 nm
Photocurrent Detected (pA)

y = mx + b
m = 0.0 +− 3.5
5 b = −0.3 +− 6.4
(Vr = 0.6V ,I = −0.3pA) 2
X =0
20 365.0 nm

0
15

−5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
10 Magnitude of Retarding Voltage Applied (V)

546.1 nm
0
FIG. 3: Graphical demonstration of the High and Low Voltage
577.0 nm
Linear Fit Method of stopping voltage determination. Calcu-
−5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
lations based on the normalized data points with error for
Retarding Voltage Applied (V) 404.7 nm wavelength light.

FIG. 2: Plot of photocurrent as a function of the magnitude of Since the asymptotic behavior of each curve at both
retarding voltage applied for each wavelength of incident light. low and high values of retarding voltages (discounting
Normalized to the zero-voltage point of the lowest intensity saturation) is linear, both sections can be fit to sepa-
wavelength. rate linear regressions. The criteria for determining how
many data points to fit on each end was simple: mini-
A back current was a noticeable effect for several wave- mum number of points required for a meaningful fit while
lengths. Back current is caused by photoelectrons liber- maintaining a reasonable χ2 . A linear fit should involve
ated from the platinum anode as a result of scattered greater than two points but maintain a χ2 of less than
 3

λ = 365.0 nm ± 2 nm
Voltage (V ± 0.1V) Current (pA ± 1pA) - All Trials Statistical Analysis
All Trials Trial 1 Trial 2 Trial 3 Trial 4 Trial 5 Mean Var. Std. Dev.
0 416 416 420 420 422 418.8 5.76 2.40
0.2 298 300 301 301 304 300.8 3.76 1.94
0.4 202 200 203 203 205 202.6 2.64 1.62
0.6 126 125 127 127 129 126.8 1.76 1.33
0.8 70 70 71 71 73 71 1.2 1.10
1 34 34 36 36 37 35.4 1.44 1.20
1.2 13 13 16 16 17 15 2.8 1.67
1.4 1 1 3 3 4 2.4 1.44 1.20
1.6 -4 -4 -2 -2 -1 -2.6 1.44 1.20
1.8 -7 -7 -5 -5 -4 -5.6 1.44 1.20
2 -8 -8 -6 -6 -6 -6.8 0.96 0.98

TABLE II: Raw data table for 365.0nm wavelength light. Statistical data is provided on the right. Standard deviations from
these measurements were used as a primary measure of random error.

100. Therefore three data points were chosen on each

end of the data set for their respective regressions.
The intersection of these two theoretical fits is a point
(Vs , I0 ). Vs is an estimate for the stopping voltage extrap-
olated from the two linear relations and I0 the baseline
current. Performing this analysis on all four sets of data,
we obtain the results tabulated below:

Wavelength (λ) St. Voltage (Vs ) Error (σVs ) Relative Error

365.0 nm 0.76 V 0.16 V 21.6%
404.7 nm 0.60 V 0.12 V 20.1%
546.1 nm 0.44 V 0.09 V 20.2%
577.0 nm 0.44 V 0.09 V 20.3%

TABLE III: Obtained by the Linear Fit Method, we find a

linear dependence of stopping voltage values on wavelength
of light.
FIG. 4: The derivative of the function Kmax (ν) gives us the
value for h/e. The χ2 for this fit is 0.21 due to the large error
bars.
Since equation 2 can be written as,
Kmax = eVs = hν − W0 , (3) 4.2. Method Two for Voltage Cut-Off
Determination: Zero-Slope Method
We can divide by e and rewrite the equation in units
of electron-Volts (eV).
Observing the qualitative character of the normalized
Kmax [in eV ] = Vs = (h/e)ν − W0 [in eV ] (4) currents vs. voltages graphs one can conceive of a slightly
Figure 5 shows a plot of the maximum determined ki- more direct way of accounting for the effects of back cur-
netic energy (in electron-Volts) of photoelectrons as a rent. If we treat (somewhat erroneously but reasonably)
function of the frequency of light (in Hz). The linear the back current as a constant negative offset on the curve
regression shows the best fit line through this set of re- in the high-voltage region of the data, where the current
duced data points. The slope of this line corresponds to should be zero, we can extrapolate the stopping voltage
the value of Planck’s constant h in eV · s. to be the point at which the curve reaches a constant
value and does not change for any larger magnitude ap-
plied voltage.
This method yields a value for h of (9.4 ± 4.8) × We call this method the Zero-Slope or Point-Deviation
10−16 eV · s. Method. It reads the data points for each wavelength
 4

from right to left and picks out the first point that devi- dissipative elements. Thus, the electron sees not the ap-
ates from the flat-line end behavior and assigns that value plied voltage V’ but a voltage V adjusted by the work
to be the stopping voltage for that particular wavelength. functions of the two metals. An illustration (Figure 6)
We tested the precision of this method by setting differ- from the same source [1] pg. 20 helps facilitate the ex-
ent threshold (minimum deviation) values and observing planation.
any change in resulting data set. For a threshold be-
tween 0 and 0.5 this method yielded the exact same set
of points and values. Since this method relies on discrete
points, the ”real” value can fall anywhere between the
chosen value and the next points on either end, so we
estimated the error in stopping voltage to be about the
interval between successive data points.

FIG. 6: The actual voltage overcome by the electron in its

journey is V = V ′ − (φa − φc ). Here, φa = Wa and φc = W0 .

If we solve for the actual y-intercept in the relation-

ship we find it to be the work function of the anode (Wa )
instead of the cathode. However, this is still not well-
represented by the data. If our objective was to obtain
an accurate work function, we would need to change cer-
tain aspects of our experiment. We would want to use
FIG. 5: The graph of Kmax vs. ν for the data set determined purer metals as opposed to alloys. Care must be taken to
by using the Zero-Slope Method. The linear fit line has a χ2
avoid cathode material deposition onto the anode. Mea-
value of 0.65.
surements should be extended until current saturation
(the voltage becomes an accelerating voltage) so we can
Using this method of analysis, h was deter- better define and correct for the back current effect.
mined to be 2.9 × 10−15 ± 7.7 × 10−16 eV · s. In this case we’re mostly interested in the slope of the
relation between Kmax and ν for our value of h/e. The
two error contributions are systematic and random error.
4.3. Combined Result The systematic error is set by the difference betweeen our
two methods of evaluating the cut-off voltage while the
The combined result of the two analyses is an random errors propagate through our analysis from their
average h value of 1.92 × 10−15 eV · s with a total origins in the measurements we took.
error of 1.08 × 10−15 eV · s. This will be derived and The random error in our raw data is determined ex-
elaborated upon in the next section. perimentally through five independent trials 1 using the
formula for the square-root of the variance, σI .
5
1X
5. ERROR CONTRIBUTIONS AND σI2 = (Ii − hIi)2 (5)
CALCULATIONS 5 i=1

Any linear regression y = mx + b calculated from our

Incidentally, according to theory, the y-intercept of the input variables and errors outputs an uncertainty on m
relation defined by Equation 4 should correspond to the
and on b. In the case of finding the intersection of two
work function (W0 ) of potassium (2.3 eV). However, we linear regression lines y = m1 x+b1 and y = m2 x+b2 , the
see that this is not nearly the case. More subtle forces error propagation formula was used to derive the follow-
are at play which contribute to the disparity between ing result for error on the X-coordinate of the intersection
theoretical predictions and actual results.
The applied voltage difference between the cathode
and the anode metals is not the voltage difference that
an electron ”sees” as it is traveling between them. Fol- 1 These trials were assumed to be independent (measured from
lowing the argument presented by Melissinos [1], energy low voltage to high voltage each time) but in some cases did not
loss around a closed circuit loop must be zero with no match the statistical distribution for independent random error.
 5

point in 2-D space: teraction with matter. We were able to demonstrate for
 2  2 light the linear dependence of its energy on its wavelength
1 1 and determine to some accuracy the constant of propor-
σx2 = σb22 + σb21 tionality. The actual value of Planck’s constant fell out-
m1 − m2 m1 − m2
 2  2 side of our errorbars, which is less than ideal. Our final
b2 − b1 2 b2 − b1 2 uncertainty on the value proved to be 50% of the value
+ σm + σm (6)
(m1 − m2 )2 1
(m1 − m2 )2 2
itself. This is very large and makes the task of drawing
solid conclusions difficult.
When faced with two different calculated values for
Planck’s constant (h1 and h2 ), and their respective un- Though we cannot necessarily reduce random errors,
certainties (σh1 and σh2 ), we determined our final value we can seek to better characterize it in our experiments.
of h by taking the average, This can be accomplished through not only more re-
h1 + h2 peated trials but especially more independent trials. If,
h = h̄ = = 1.92 × 10−15 eV · s (7) in the future, we attempt to reset the amp meter and
2
voltage supply for each trial or wait a longer period of
and determined our random error on h, σhr , by prop- time between trials, we may find a more representative
agating both uncertainties using the formula, variance in our measurements.

1 2 1 There are means, however, of reducing our systematic

σh2 r = σ + σ 2 = 2.12 × 10−31 eV 2 s2 . (8)
4 h1 4 h2 error. In addition to the changes and improvements sug-
gested previously at the end of the discussion on work
The systematic error was estimated by the square-root
functions, by simply taking more data points and ex-
of the variance of h1 and h2 and determined to be 9.8 ×
tending our observations deeper into the high and low
10−16 eV · s.
voltage ends, we avail ourselves of more discriminating
The total error on h is a combination of random error
information to adjust for our systematic error. Further-
(σhr ) and systematic error (σhs ),
more, the usage of a brighter source can offer us more
σh2 = σh2 s + σh2 r (9) resolution on our curves and allow us to better identify
a cut-off voltage.
σh = 1.08 × 10 −15
eV · s
There are extensions on our experiment that may and
should be explored in the future. Changing other vari-
6. CONCLUSIONS ables, including intensity and the cathode/anode work
function and observing its effects on the system, may
Our experiment verified our hypothesis. We observed provide us with further and confirming evidence on the
light behave as if it were a particle (not a wave) in its in- particle nature of light.

[1] Melissinos, A.C., Experiments in Modern Physics - 1st

Edition, Academic Press, [1966]
[2] Bevington and Robinson, Data Reduction and Error Anal-
ysis for the Physical Sciences - 3rd Edition, McGraw-Hill,
[2003]

Acknowledgments

We are grateful of the assistance we received from the

Junior Lab TA’s and Professor with this experiment.
All linear regressions were calculated using the fitlin.m
MATLAB program offered on the Junior Lab website
http://web.mit.edu/8.13/www/jlmatlab.shtml.