Biochem. J.

(1987) 248, 677-682 (Printed in Great Britain) 677

A stopped-flow calorimeter for biochemical applications

J. Victor HOWARTH,*t Neil C. MILLAR and Herbert GUTFREUNDt
*The Marine Biological Association, Citadel Hill, Plymouth PLI 2PB, U.K., and tDepartment of Biochemistry,
University of Bristol, Bristol BS8 ITD, U.K.

A rapid-response stopped-flow calorimeter for small samples of reagents is described. The construction,
performance characteristics and operational limitations are described, along with an example of its ability
to resolve the kinetics of an enzyme-catalysed hydrolysis. It is thought likely that the method would find
useful application in a variety of chemical and biochemical investigations.

INTRODUCTION performing the reaction in different buffers having high

The calorimeter described in this paper has been and low heats of ionization. This procedure has been
developed for the study of rapid heat changes in solution, successfully used by Millar et al. (1987) in a study of
such as accompany many enzyme reactions. The use of myosin ATPase.
heat, or temperature, changes to monitor rapid chemical For such measurements the requirement is for an
reactions dates back to some of the earliest experiments instrument having a time resolution of a few milliseconds
with continuous-flow techniques (Roughton, 1930). The and a sensitivity in the range < 1 mK. This is similar to
success of Roughton's investigations on the kinetics of
the specification required for measurement of heat
haemoglobin-oxygen binding proved the value of such changes in active nerve and muscle cells (Howarth et al.,
thermal kinetic studies, but also revealed the principal 1979; Curtin et al., 1983). In the present paper we
shortcoming of continuous-flow systems, namely that describe such an instrument, together with tests of its
they are very extravagant in the use of reactants. performance characteristics and some features of its
Kodama & Woledge (1985) described two batch practical use. In the accompanying paper (Millar et al.,
calorimeters that they designed and used for heat studies 1987) we report studies of the resolution of substrate
of different enzyme reactions. These required reactant binding, chemical catalysis and product dissociation
volumes of only 1 ml and had a time resolution of 1 s or during the reaction of myosin with ATP, results that
further demonstrate the potential of the technique.
less. It seems likely that in these instruments Kodama &
Woledge have pressed the development of the batch
calorimeter close to its limit of speed, and thus, for still DESCRIPTION OF THE INSTRUMENT
better time resolution, a different system is called for. A
stopped-flow system would seem to be a promising Design considerations
option. Compared with continuous-flow systems, the The instrument has essentially the same configuration
stopped-flow technique is much more economical in the as an optical stopped-flow device but with a thermal
use of reagents and it has been successfully used for the detector positioned immediately downstream from the
study of transient kinetics by a variety of optical mixing chamber at the point where optical changes are
measurements. Consequently, there have been a number normally observed. For the detection of thermal changes
of proposals and attempts to adapt the stopped-flow physical requirements are most easily seen if the system
method for the study of rapid enthalpy changes. is though of as being in two parts. The first part
Kodama & Kometani (1986) used this method in a trial comprises the two columns of fluid containing the
study of ATP hydrolysis by myosin subfragment 1 but reactants upstream from the mixing chamber. It is
give only a preliminary outline description of the essential that these are at precisely the same temperature
instrument. Other earlier essays in this field are reviewed and that there are no persistent thermal gradients along
by Berger et al. (1984), but, although the later instruments their length. This is best achieved if both are in contact
described there appear to have considerable promise, with a large thermal mass of high thermal conductivity.
they do not seem to have been much exploited for the The second part comprises the reacting solution in the
study of enzyme kinetics. mixing chamber and the reaction chamber. Here the
That such studies are likely to be rewarding is indicated requirement is that the walls of the chamber and the
by previous experience of investigating the same reaction detector itself should abstract the least possible heat
by different transient kinetic techniques (Gutfreund, from the reacting solution and that heat should be lost as
1975). The non-specific nature of temperature measure- slowly as possible to the environment. This requires that
ments, although a drawback in some applications, can be the reacting solution must be in contact with a very small
turned to advantage in certain reaction systems by thermal mass of low thermal conductivity. Thus it is seen
providing a means of monitoring those steps in the that the two parts of the system have exactly the opposite
reaction sequence that are spectroscopically silent. requirements. These are satisfied in our apparatus by
Moreover, a powerful additional tool is presented by arranging for the initial fluid columns to be contained in
time-resolved heat measurements, namely reaction steps narrow-bore syringes and 2 mm-bore delivery tubes that
involving ionizations can be specifically characterized by are in direct contact with a large-volume water thermo-

$ To whom correspondence should be addressed.

Vol. 248
678 J. V. Howarth, N. C. Millar and H. Gutfreund

.... -,............. ......

.......

Pushing~~~~~~~~~~~~~~~~.
r~ ~ ~ ~ ~ ~ ~ ~ ~ ~ .
/~ ~ ~ ~ . . . . . .
syrnge ...............

Trigger
cells

Li

Fig. 1. Diagram showing lay-out of the stopped-flow calorimeter

The compressed air tubes to the pneumatic cylinders and leads to the thermopiles and trigger are omitted for clarity.

stat, and for the second part, by using for the reaction both directions of movement by needle valves restricting
chamber, a very-thin-walled tube insulated from the the exhaust side. The excursion of the main pushing
thermostat by an air jacket. This led to the configuration cylinder is restricted by two pieces of aluminium channel
as shown in Fig. 1. clipped over the through-rod at the front and rear ends.
The instrument described here is the one presently in The rear one carries a piezo-electric wafer, sealed into the
use and all of the experiments described in the accom- cut-offend of a plastic disposable syringe, which provides
panying paper (Millar et al., 1987) were made with this a trigger signal at the precise moment when the rod hits
instrument. the stop. Movement of the piston can be monitored by a
conducting plastic linear transducer that is also coupled
Features of the instrument to the rear extension of the through-rod but not shown
In use the whole assembly on its base-plate is immersed in the diagram.
in a well-stirred 80-litre water bath in a heavily insulated In this arrangement the flow is arrested upstream from
box. All operation of the instrument is from outside the the calorimeter, and after each push the delivery syringes
box by remote control. Except at 0 °C, when there is ice are re-charged with a single portion of reactant solution
present, no attempt is made to regulate the temperature (usually 150 1,) ready for the next push. More commonly
of the bath. Rather, it is adjusted initially and then in stopped-flow systems the flow is arrested by stopping
allowed to drift. In the course of a 4-5 h experiment the a collecting syringe on the downstream side of the
bath never drifts more than 0.1 K and then so slowly that observation cell. This method was not open to us because
the reference baseline for transient measurements is not it leads to a large pressure jump at the moment of stop,
affected. which would cause a large temperature change due to
Fig. 1 shows the instrument as it is normally used, with thermo-elastic heat released in the fluid (and more than
two calorimeters operated differentially. The valve block likely rupture the thin walls of the observation tube).
is made from aluminium alloy with a slot to take a close- Bowen et al. (1980) describe another solution to this
fitting polytetrafluoroethylene slider, which is drilled to problem using a push-pull flow arrangement. We have
form a four-ganged three-way tap so that in one position not tried this but it could be added to this instrument and
the delivery syringes are in straight-through communica- may be better than the simple upstream stop.
tion with the delivery tubes and thence to the calorimeter
cells, and in the other position the delivery syringes Calorimeter cells
connect to the storage syringes for re-charging. Actuation A calorimeter cell of the current type is illustrated in
of the slider is by the two air cylinders. The third air Fig. 2. It has four parts: a mixing chamber, a reaction
cylinder actuates the delivery syringes. All three air chamber, a thermopile and an air jacket.
cylinders are 25 mm-bore double-acting cylinders and
are supplied with air at 0.8 MPa (8 bar) pressure from Mixing chamber. The requirement of the mixing
outside the box via nylon tubes (which are not shown in chamber is that it provides for rapid and complete
the illustration). Speed of the cylinders is controlled in mixing but with the least amount of mechanical work,
1987
A stopped-flow calorimeter 679

(a) _ 1 (b) (c)

IJ i~~~h
rc

.i.. m1.d.. ..... ... .... ... ..

....

mc
Fig. 2. Diagram of a calorimeter cell showing the mixing chamber and the reaction chamber in section
(a) A single calorimeter cell. Key: th, thermopile; mc, mixing chamber; rc, reaction chamber; dt, delivery tubes. The two
reactant solutions enter the mixing chamber at the sides and exit via the thin-walled tube that forms the reaction chamber. The
thickness of the thermopile is exaggerated in relation to the diameter of the reaction chamber. (b) Cross-section through the
thermopile at the level Y-Y. (c) Cross-section through the mixing chamber at the level X-X.

and hence heat, put into the fluid. We do not know of force of constantan/chromel couples is 59.1 uV * K-' at
any established theoretical treatment that would provide 273 K. Thermopiles made in this way give the full
a basis for such a design, so the form and dimensions of theoretical output provided that none of the couples is
the present chambers have evolved by intuition rather short-circuited. The thermopile is inserted through slits
than by design. Mixing chambers may be made by cut in the side of the tube, arranged so that the 'hot'
drilling in Perspex [poly(methyl methacrylate)]. This is junctions lie along the midline, and is then sealed in with
satisfactory for single cells, but for paired cells to be used a small amount of epoxy resin.
in a differential mode it is important to have the two
chambers exactly alike and it is difficult to machine them Air jacket. The air jacket is simply the end cut from a
with the necessary precision. The two mixing chambers 5 ml plastic disposable syringe. The thermopile passes
in the present cells were made by casting liquid epoxy through, and is sealed into, slits cut in the wall so that the
resin, both in the same mould. This method has given the electrically insulated 'cold' junctions protrude into the
best results to date. It is important to ensure that the bath water.
reaction chamber tube is truly central where it projects
into the mixing chamber and that the gap between the Characteristics of the instrument
end of the tube and the back of the mixing chamber is Three characteristics determine the usefulness of the
exactly the same in the two cells. It is our practice to view instrument in a particular application. They are sensi-
the mixing chambers under 20 x magnification and to tivity, speed of response and rate of heat loss.
remove surface irregularities at this level.
Sensitivity. Fig. 3(a) shows a series of records obtained
Reaction chamber. The reaction chamber, where the by reaction of different concentrations of HCl with Tris
measurements are made, is a tube that carries the buffer at pH 8. In Fig. 3(b) the output voltage is plotted
reaction solution from the mixing chamber. The require- against temperature change calculated from the known
ments are that the heat capacity of the wall must be small enthalpy change of the reaction. The sensitivity is given
compared with that of the contained fluid and that the by the slope of the regression line, giving a value of
wall must come rapidly into temperature equilibrium 5.95 mV K-', exactly the same as the expected theoretical
-

with the fluid. Both of these are met by having a thin- value for 100 thermojunctions. This procedure gives the
waIled tube. Our tubes are made by spiral wrapping a true temperature sensitivity of the instrument, but
tape, cut from 10,tm polyester film, around a 2 mm (unlike a batch calorimeter) it slightly overestimates the
polythene former, the width of the tape and the pitch of heat calibration because it does not take account of
the spiral being calculated to give a double thickness of the heat capacity of the thermopile and the wall of the
film at all points along the wall. The tape is coated with reaction chamber. This is because the neutralization
epoxy resin as it is wound and excess resin is removed. reaction is so quick that all the heat is released before the
When the resin has hardened the polythene former is flowing solution reaches the thermopile. Typically the
withdrawn. These tubes weigh 2.37 mg for 5 mm length flow continues for 30 ms or more, giving plenty of time
(i.e. the length of the thermopile) and have a heat for the thermopile and wall to come into temperature
capacity of about 2.76 mJ * K-1, or about 4 % that of the equilibrium. Thus, when the flow finally stops, the fluid
contained solution. in contact with the thermopile does not lose any heat to
these 'inert' elements. On the other hand, the enzyme
Thermopile. The thermopiles are made from 25 ,m- reactions that the instrument is designed to investigate
diameter constantan and chromel wire. The wires are are not so fast and have scarcely got underway before the
embedded in epoxy resin and insulated by sandwiching solution reaches the thermopile after mixing. Thus, when
between two layers of 10 ,um polyester film. The the flow is arrested, all subsequent reaction heat is shared
thermopile consists of 100 couples in series, has an with the inert elements, leading, in the present instrument,
electrical resistance of 2300 Q and extends 5 mm along to a 5 0 reduction in temperature change.
the line of flow. The heat capacity of the thermopile in
contact with the solution is 0.8 mJ K-1, which is close to
-
Speed of response. Fig. 4 shows a record obtained by
1 % of that of the solution. The thermo-electromotive passing a 2 ms pulse of current between electrodes placed
Vol. 248
680 J. V. Howarth, N. C. Millar and H. Gutfreund

61o (a)

41

20

0 -
4(U
Q
-1.0 -0.5 0 0.5 1.0 0 10 20 30
Time (ms)
I-
Time (s)
Fig. 4. Response time of the calorimeter
L..
Q
A 2 ms pulse of heat (marked by the horizontal bar) was
60 1_ (b) produced by passing an electric current pulse through an
electrolyte solution in the reaction chamber. This is the
average of six records with the pulse polarity reversed
40 _ 00-0 between successive records. Time zero marks the time at
which the pulse starts. The fitted exponential is fitted from
2 ms and has a rate constant of 302 s-'.
20 0

0 1 1

0 0.25 0.50 0.75 1.00

[HCI] (mM)
Fig. 3. Calibration with Tris versus Ha (U

(a) Calorimeter records corresponding to the eight HCI E

concentrations shown in (b). Time zero marks the time at a)

which flow stops. (b) Peak thermopile output versus HCI .91

concentration. Slope of fitted line = 67.2 #uV m-'. Since

AH for ionization of Tris = -47.5 kJ mol-1, this corres-
ponds to 5.95 mV K-1. Conditions: 25 mM-Tris, [HCI] as
indicated (concentrations refer to solution in reaction 0 100 200 300
chamber after mixing), pH 8.0, 5 'C. Time (s)
Fig. 5. Cooling curve showing rate of heat loss from th4
calorimeter
in the column of fluid a few millimetres away from the
thermopile at each end. At the beginning and end of the This is one of the records from the experiment illustrated
pulse the record is disturbed by capacitive coupling of in Fig. 3 but on a much slower time scale. Conditions:
the pulse into the thermopile circuit, but it can be seen 250 /,M-HCl versus 25 mM-Tris, pH 8.0, 5 'C. Time zero
that the record takes off during the 2 ms and at the end marks the time at which flow stops. The fitted exponential
has a rate constant of 0.0168 s-'.
of the pulse the record continues to its maximum value in
an exponential fashion with a time constant of 4.5 ms.
The calorimeter itself is thus very fast. In normal use the
effective speed of response of the system is determined by there are a number of practical factors that need always
the bandwidth set in the amplifier to achieve an acceptable to be borne in mind. Some of these are outlined
signal-to-noise ratio. below.
Rate of heat loss. Heat released into the reaction Work heat. Considerable mechanical work is per.
chamber escapes to the bath by three, approximately formed on the solutions as they are driven through the
equal, parallel routes: upstream and downstream along delivery tubes and the mixing chambers, resulting in the
the fluid column and laterally via the conductors of the release of heat. Fig. 6 shows a record of this work heal
thermopile. Heat loss across the air space is insignificant. resulting from driving water alone through the calori-
Fig. 5 shows a cooling curve and demonstrates that it is meter. Much of the development work has been aimed al
fitted well by a single exponential having a rate constant minimizing this work heat, but some is unavoidable and
of 0.017 s-' (time constant 58.8 s). This remains the same therefore has to be treated as the baseline of measure-
in different experiments, and enables the heat loss to be ment. When a reaction is to be measured the heat due te
corrected with sufficient precision to allow observations the reaction can be observed by subtracting a control
to be made over many minutes. trace obtained with solutions that mimic the physical
properties of the reacting solutions, in particular the
The instrument in use viscosity and concentration, but with no reaction present.
The above specification characteristics meet most of There are two ways to subtract the control traces. In
the requirements of the original project. In use, however, the first, the control records are made in separate pushes
1987
A stopped-flow calorimeter 681

to use and if something goes wrong in the course of an

experiment it is impossible to know which of the two
sides is at fault.
It is, of course, important to minimize the work heat as
far as possible by the design of the system. It is of
obvious advantage to preserve laminar flow in the
delivery tubes. For this they must have a smooth interior
surface and the flow should not be subjected to rapid
changes of diameter or direction. Their dimensions are
arranged to keep the Reynolds number at the expected
flow rates in the region below 2300, the critical value
I.U above which laminar flow becomes unstable. The delivery
Time (s)
tubes between the valve and the mixing chambers hold
E
somewhat more fluid than is ejected in a single push. This
means that fluid that may have been heated by being
forced out of the syringes and through the valve does not
reach the mixing chambers until the following push. The
half-time of temperature equilibration between the tubes
and the bath is calculated to be 1.7 s, so that there is
plenty of time in the interval between pushes (usually
8 min) for any temperature gradients in the tubes to be
dissipated.
Flow rate. Work heat is very sensitive to flow rate, and
to keep the former to convenient levels it is necessary to
0
operate the instrument at flow rates considerably less
Time (s)
than is usual in optical stopped-flow systems. However,
if the flow is too slow problems may arise with mixing
Fig. 6. Records of work heat and scouring.
(a) Output from a single calorimeter cell due to water Mixing was tested at different speeds by using acid-
versus water. Time zero is the time at which flow stops. base reactions to obtain records such as those of Fig.
The change in level between the baseline and the horizontal 3(a). There it can be seen that no heat is produced after
part of the record following the push is the heat produced the stop, as it would be if incompletely mixed solutions
by the mechanical work done on the fluid to overcome continued to react by diffusion in the stationary fluid in
viscous resistance. (b) The difference signal seen at higher the reaction chamber.
gain with the use of the paired calorimeter cells, i.e. (water Scouring is the term used in the laboratory to refer to
versus water) (water versus water). The negative spike is
- the displacement of the old fluid in the reaction chamber
the result of imperfect symmetry between the two halves of by the incoming fluid during the push. This was also
the system, but the work heat is cancelled 25 ms after the tested by using the acid-base reaction, by making a series
stop. This is the average of three traces. Conditions: flow of pushes in quick succession. If scouring is complete the
velocity 2 m/s, distilled water in all four syringes, 5 'C. records all peak at the same level, but if there is
incomplete scouring the first record is low because of the
extra heat capacity of the remnants of the old solution.
The second record, if made before there has been
before or after the reaction records. This needs only one appreciable heat loss, goes higher, and so on in a
calorimeter but it requires that the work heat must be the diminishing staircase to the final value.
same in all pushes. In our system this is not found to be By these tests we found that a flow rate of 2 ms-' in the
the case. The variation is probably largely due to reaction chamber and a push volume of 300 ,u (150 #1
inconsistency in the speed of operation of the air from each syringe) give good results.
cylinders. These cylinders are not designed for precise
control of speed, and it is likely that they would be Push artifacts. In addition to the work heat, Fig. 6 also
improved by hydraulic damping or, perhaps better, they displays a spike-like artifact during the push. This
could be replaced by hydraulic actuators. artifact has manifested itself in all versions of the
In the second method a second calorimeter is used to instrument, and although other artifacts have been
record the control trace at the same time as the reaction identified and eradicated this one has so far proved
trace. This method avoids the problem of variation in resistant to treatment. Although it is mostly cancelled in
pushing speed, but it requires that the two calorimeters the subtraction, there is invariably some small residual,
are exactly alike in their characteristics. This is the due to imperfect symmetry, which persists for a short
method that has yielded the best results in our experi- time after the stop and may interfere with measurements
ments (see Fig. 6b). in the first 20 ms. For reactions lasting longer than this
The most obvious and elegant method would be to use it does not get in the way. The relative magnitude of the
a truly differential system by placing the opposite ends of artifact, the difference signal and a reaction trace under
a thermopile in a pair of parallel reaction tubes in the typical experimental conditions can be seen by comparing
same enclosure, one carrying the reaction mixture and Figs. 6 and 7.
the other the blank. We have made some good recordings We are not altogether sure as to the cause of this
using this system, but the arrangement is not at all easy artifact. Various different tests have shown that it is not
Vol. 248
682 J. V. Howarth, N. C. Millar and H. Gutfreund

where AT, = temperature change at time t, At = time

increment and k = rate constant of heat loss. Analysis of
the recorded traces was done in the computer using
standard non-linear fitting programs- (Edsall & Gut-
freund, 1983).
E
E L / 10.5 mK CONCLUSION
Fig. 7 shows an experimental record of the time course
of heat production during a single turnover of the
0 100 200 300 enzymic hydrolysis of acetyltyrosine ethyl ester by
Time (ms) chymotrypsin. The fitted exponential gives a rate constant
Fig. 7. Hydrolysis of acetyltyrosine ethyl ester by chymotrypsin close to 20 s-' for the reaction under the stated conditions.
It is clear from this that the calorimeter in its present
Conditions: 250 mM-chymotrypsin, 90 /tM-acetyltyrosine form is a practical working research instrument, and
ethyl ester, 0.1 M-imidazole buffer (reaction chamber further examples of its application are given in the
concentrations), pH 7.0, 22 'C. Time zero marks the time accompanying paper (Millar et al., 1987). At the slow
at which flow stops. The observed temperature rise is due end of its range it reaches and overlaps well into the
to the release of 1 mol of H+/mol of acetyltyrosine ethyl range where batch calorimeters become a better choice
ester as the acyl-enzyme complex breaks down to form free
acetyltyrosine and enzyme. The fitted exponential has a (Kodama & Woledge, 1985), and at the quick end it
rate constant of 19.6 s-1 and an amplitude 0.79 mK, which extends the useful range of thermal kinetic studies in vitro
corresponds to an enthalpy change of -37 kJ mol-'.
- into a hitherto inaccessible region. Because the develop-
ment of the device has been associated with a research
project we have not been able to take time to optimize
systematically all of the components of the system but
due to rapid heat loss to some inert heat capacity such as have instead progressed by experience and intuitive short
the walls of the chamber, nor is it caused by distortion of cuts using components and materials that came readily
the thin-walled tubes. Also, it is not an electrical artifact to hand. Further development of the system should
such as might be induced in the thermopile circuit certainly be possible.
electro-magnetically or by some form of charge separa-
tion during the flow. A special 'calorimeter' having an
element made exactly as a themopile save that all its
conductors were of constantan (instead of constantan-
chromel couples), and therefore insensitive to temper- This work is part of a project generously supported by The
ature, did not show any sign of this artifact. Therefore Wellcome Foundation through a Post Doctoral grant for
the artifact must be a temperature, and hence a heat, N. C. M. Other funds were kindly provided by the Science and
change. The falling phase after the stop is many times Engineering Research Council and the Muscular Dystrophy
faster than the rate of heat loss from the system and Association. We are also much indebted to Mr. Percy Austin
and Mr. Dave Purse for their skills and interest.
therefore must be due to heat absorption. This, and some
other properties of the artifact, suggest that it is caused,
at least in part, by pressure changes. If it is due to
thermoelastic heat in the fluid it would require pressure
changes of the order of I atm to account for the observed
magnitude of the artifact. REFERENCES
Berger, R. L., Clem, T. R., Harden, V. A. & Mangum, B. W.
Data acquisition. The output from each calorimeter is (1984) Methods Biochem. Anal. 30, 269-329
in the region of I #tV and is amplified by a low-noise d.c. Bowen, P., Balko, B., Blevins, K., Berger, R. L. & Hopkins,
differential instrumentation amplifier, based on the AD- H. P. (1980) Anal. Biochem. 102, 434-40
524-AD, followed by an OP-07C, giving an overall gain Curtin, N. A., Howarth, J. V. & Woledge, R. C. (1983) J.
of 105. The amplifier suffers no discernible baseline drift Muscle Res. Cell Motil. 4, 207-222
over a period of several minutes, long enough for the Edsall, J. T. & Gutfreund, H. (1983) Biothermodynamics, pp.
slowest observations ever likely to be made with this type 228-236, John Wiley and Sons, New York
of calorimeter. The amplified signals from the two Gutfreund, H. (1975) Prog. Biophys. 29, 161-173
channels, or difference signal as required, are taken to a Howarth, J. V., Ritchie, J. M. & Stagg, D. (1979) Proc. R. Soc.
Nicolet 3091 digital oscilloscope, and the output, London B 205, 347-367
digitized as 4000 12-bit points, is displayed and stored on Kodama, T. & Kometani, K. (1986) Proc. Jpn. Acad. Ser. B
an Apple lIe computer. In records of short duration heat 62, 105-108
loss is insignificant, but for recordings exceeding about Kodama, T. & Woledge, R. C. (1985) Rev. Sci. Instrum. 56,
10 s it is corrected in the computer according to the 40-42
relation: Millar, N. C., Howarth, J. V. & Gutfreund, H. (1987) Biochem.
0 J. 248, 683-690
'ATt(corrected) AT(observed) + t((bsreobserved)
E [AT7(b I(1 e kAt)] Roughton, F. J. W. (1930) Proc. R. Soc. London A 126,
439-486

Received 3 April 1987/16 June 1987; accepted 14 August 1987

1987