Hilbert Space Models Commodity Exchanges

Paul Cockshott
Dept Computing Science University of Glasgow
wpc@dcs.gla.ac.uk
Abstract. It is argued that the vector space measures used to mea-
sure closeness of market prices to predictors for market prices are invalid
because of the observed metric of commodity space. An alternative rep-
resentation in Hilbert space within which such measures do apply is
proposed. It is shown that commodity exchanges can be modeled by the
application of unitary operators to this space.
1 Linear Price Models
The context of this paper is the empirical testing of linear models of eco-
nomic activity. Whilst these originated in an informal way in the work of
Adam Smith and Quesney, and were partially formalised by Marx in volume
3 of Capital, an adequate formal treatment had to wait for von Neuman[21]
and Kantorovich[9]. Both von Neumann and Kantorovich were mathematicians
rather than economists. Their contributions to economics were just one part of
a variety of research achievements. In both cases this included stints working on
early nuclear weapons programs, for the US and USSR[15] respectively. At least
in von Neumanns case the connection of his economic work to atomic physics
was more than incidental. One of his great achievements was his mathematical
formalization of quantum mechanics[22] which unied the matrix mechanics of
Heisenberg with the wave mechanics of Schrodinger. His work on quantum me-
chanics coincided with the rst draft of his economic growth model[21] given as a
lecture in Princeton in 1932. In both elds he employs vector spaces and matrix
operators over vector spaces, complex vector spaces in the quantum mechanical
case, and real vector spaces in the growth model. Kurz and Salvadori [11]argue
that his growth model has to be seen as a response to the prior work of the
mathematician Remak[14], who worked on superposed prices.
Remak then constructs superposed prices for an economic system
in stationary conditions in which there are as many single-product pro-
cesses of production as there are products, and each process or product
is represented by a dierent person or rather activity or industry. The
amounts of the dierent commodities acquired by a person over a cer-
tain period of time in exchange for his or her own product are of course
the amounts needed as means of production to produce this product
and the amounts of consumption goods in support of the person (and
P. Bruza et al. (Eds.): QI 2009, LNAI 5494, pp. 299307, 2009.
c Springer-Verlag Berlin Heidelberg 2009
300 P. Cockshott
his or her family), given the levels of sustenance. With an appropri-
ate choice of units, the resulting system of superposed prices can be
written as
p
T
= p
T
C
where C is the augmented matrix of inputs per unit of output, and
p is the vector of exchange ratios. Discussing system Remak arrived
at the conclusion that there is a solution to it, which is semipositive
and unique except for a scale factor. The system refers to a kind of
ideal economy with independent producers, no wage labour and hence
no prots. However, in Remaks view it can also be interpreted as a
socialist economic system [11].
With Remak the mathematical links to the then emerging matrix mechanics are
striking - the language of superposition, the use of a unitary matrix operator
C analogous to the Hermitian operators in quantum mechanics
1
. Remak shows
for the rst time how, starting from an in-natura description of the conditions
of production, one can derive an equilibrium system of prices. This implies that
the in-natura system contains the information necessary for the prices and that
the prices are a projection of the in-natura system onto a lower dimensional
space
2
. If that is the case, then any calculations that can be done with the
information in the reduced system p could in principle be done, by some other
algorithmic procedure starting from C. Remak expresses condence that with
the development of electric calculating machines, the required large systems of
linear equations will be solvable.
The weakness of Remaks analysis is that it was limited to an economy in
steady state.Von Neumann took the analysis on in two distinct ways:
1. He models an economy in growth, not a static economy. He assumes an econ-
omy in uniform proportionate growth. He explicitly abjures considering the
eects of restricted natural resources or labour supply, assuming instead that
the labour supply can be extended to accommodate growth. This is perhaps
not unrealistic as a picture of an economy undergoing rapid industrialization
( for instance Soviet Russia at the time he was writing ).
2. He allows for there to be multiple techniques to produce any given good -
Remak only allowed one. These dierent possible productive techniques use
dierent mixtures of inputs, and only some of them will be viable.
von Neumann again uses the idea of a technology matrix introduced by Re-
mak, but now splits it into two matrices A which represents the goods con-
1
Like the Hermitian operators in quantum mechanics, Remaks production operator
is unitary because pis an eigen vector of C and |p| is unchanged under the operation.
2
Suppose C is an n n square matrix, and p an n dimensional vector. By applying
Iversons reshaping[8,7] operator , we can map C to a vector of length n
2
thus
c (nn)C , and we thus see that the price system, having ndimensions involves
a massive dimension reduction from the n
2
dimensional vector c.
Hilbert Space Models Commodity Exchanges 301
sumed in production, and B which represents the goods produced. So a
ij
is the
amount of the j th product used in production process i, and b
ij
the amount
of product j produced in process i. This formulation allows for joint produc-
tion, and he says that the depreciation of capital goods can be modeled in this
way, a production process uses up new machines and produces as a side ef-
fect older, worn machines. The number of processes does not need to equal the
number of distinct product types, so we are not necessarily dealing with square
matrices.
Like Remak he assumes that there exists a price vector y but also an intensity
vector x which measures the intensity with which any given production process is
operated. Later the same formulation was used by Kantorovich. Two remaining
variables and measure the interest rate and the rate of growth of the economy
respectively.
He makes two additional assumptions. First is that there are no prots,
by which he means that all production processes with positive intensity return
exactly the rate of interest. He only counts as prot, earning a return above the
rate of interest. This also means that no processes are run at a loss ( returning
less than ). His second assumption is that any product produced in excessive
quantity has a zero price.
He goes on to show that in this system there is an equilibrium state in which
there is a unique growth rate = and denite set of intensities and prices.
The intensities and prices are simultaneously determined.
Von Neumanns work was inuential in economic theory, spawing a number
of similar models, probably the most famous of which was Sraas[18].
At the time that von Neumann was writing, despite Remaks optimism, it
was not possible to empirically test dierent linear theories of prices because of
the problem of collecting the necessary data, and the problem of solving large
matrix equations. From the 1950s onwards though, the empirical problem of
obtaining the A and B matrices was reduced by the publication of national
input output tables. Since the ready availability of computers emprical testing
became possible.
In 1983 Farjoun and Machover published a seminal work applying statistical
mechanics to the dynamics of capitalist economies[6]. One of the predictions of
their book was that what they called vertically integrated labour coecients
would be good predictors for market prices. One can view their price model
as being similar to that of Remak with added thermal noise. Their predictions
have largely been born out by subsequent empirical studies [2,19,12,3,13,17,4],
though there have been isolated studies questioning this [10,20]. There has been
some controversy as to what metric was appropriate for determining the close-
ness of market prices to integrated labour coecients. In the recent literature
discussing this [10,13,16,20]it has been taken as given that the use of vector
space measures is appropriate. For example one measure proposed has been to
determine the angle between two price vectors. I wish to point out that this
approach is questionable.
302 P. Cockshott
2 The Vector Space Problem
Vector spaces are a subclass of metric space. A metric space is characterized by
a positive real valued metric function (p, q) giving the distance between two
points, p, q. This distance function must satisfy the triangle inequality (p, q)
(p, r) + (q, r). In vector spaces this metric takes the form:
(p, q) =
_

(p
i
q
i
)
2
(1)
We have argued elsewhere[1] that the metric of commodity space does not
take this form. Let us recapitulate the argument.
Conjecture 1. Commodity space is a vector space.
Assume that we have a commodity space made up of two commodities, gold
and corn and that 1oz gold exchanges for 100 bushels of corn. We can represent
any agents holding of the two commodities by a 2 dimensional vector c with
c
0
being their gold holding and c
1
being their corn holding. Given the exchange
ratio above, we can assume that (1,0) and (0,100) are points of equal worth and
assuming that commodity space is a vector space thus
((0, 0), (1, 0)) = ((0, 0), (0, 100)) (2)
This obviously does not meet equation 1 but if we re-normalise the corn axis
by dividing by its price in gold, we get a metric

c
(p, q) =
_
(p
0
q
0
)
2
+ (
p
1
q
1
100
)
2
(3)
which meets the equation we want for our two extreme points:

c
((0, 0), (1, 0)) =
c
((0, 0), (0, 1)) (4)
If this is our metric, then we can dene a set of commodity holdings that are
the same distance from the origin as holding 1oz of gold. Let us term this U the
unit circle in commodity space:
U = {a U :
c
((0, 0), a) = 1} (5)
Since these points are equidistant from the origin, where the agent holds
nothing, they must be positions of equal worth, and that movements along this
path must not alter the net worth of the agent. Let us consider a point on U,
where the agent holds
1

2
oz gold and
100

2
bushels of corn.
Would this in reality be a point of equal worth to holding 1 oz of gold?
No, since the agent could trade their
100

2
bushels of corn for a further
1

2
oz
gold and end up with

2oz> 1oz of gold. Thus there exists a point on U that
is not equidistant from the origin, hence equation 3 can not be the form of the
metric of commodity space and thus conjecture 1 falls, and commodity space is
not a vector space.
Hilbert Space Models Commodity Exchanges 303
3 The Metric of Commodity Space
The metric actually observed in the space of bundles of commodities is:

b
(p, q) =

i
[p
i
q
i
]

(6)
where p, q are vectors of commodities, and
i
are relative values. The unit
circle in this space actually corresponds to a pair of parallel hyperplanes on
above and one below the origin. One such hyperplane is the set of all commodity
combinations of positive value 1 and the other, the set of all commodity combi-
nations of value -1. The latter corresponds to agents with negative worth, i.e.,
net debtors.
Because of its metric, this space is not a vector space and it is questionable
whether measures of similarity based on vector space metrics are appropriate
for it. However it is possible to posit an underlying linear vector space of which
commodity space is a representation.
4 Commodity Amplitude Space
We will now develop the concept of an underlying space, commodity amplitude
space, which can model commodity exchanges and the formation of debt. Unlike
commodity space itself, this space, is a true vector space whose evolution can
be modeled by the application of linear operators. The relationship between
commodity amplitude space and observed holdings of commodities by agents is
analogous to that between amplitudes and observables in quantum theory.
Let us consider a system of n agents and m commodities, and represent the
state of this system at an instance in time by a complex matrix A, where a
ij
represents the amplitude of agent i in commodity j. The actual value of the
holding of commodity j by agent i , we denote by h
ij
an element of the holding
matrix H. This is related to a
ij
by the equation a
ij
=
_
h
ij
.
4.1 Commodity Exchanges
We can represent the process of commodity exchange by the application of ro-
tation operators to A. An agent can change the amplitudes of their holdings of
dierent commodities by a rotation in amplitude space. Thus an initial ampli-
tude of 1 in gold space by an agent can be transformed into an amplitude of 1 in
corn space by a rotation of

2
. Borrowing Dirac notation we can write these as
1gold, and 1corn. A rotation of

4
on the other hand would move an agent
from a pure state 1gold to a superposition of states
1

2
|gold> +
1

2
|corn .
Unlike rotation operators in commodity space this is value conserving since on
squaring we nd their assets are now
1
2
gold +
1
2
corn.
The second conservation law that has to be maintained in exchange is con-
servation of the value of each individual commodity, there must be no more or
less of any commodity after the exchange than there was before. This can be
304 P. Cockshott
modeled by constraining the evolution operators on commodity amplitude space
to be such that they simultaneously perform a rotation on rows and columns of
the matrix A.
Suppose we start in state:
A =
_
1 0
0 2
_
, H =
_
1 0
0 4
_
Where agent zero has 1 of gold and no corn, and agent one has no gold and
4 of corn. We can model the purchase of 1 of corn by agent zero from agent one
by the evolution of A to:
A2 =
_
0 1
1

3
_
which corresponds to nal holdings of:
H2 =
_
0 1
1 3
_
Note that the operation on amplitude space is a length preserving rotation
on both the rows and the columns. The lengths of the row zero and column zero
in A2 are 1 the lengths of row and column one is 2 just as it was for A. This
operation can be eected by the application of an appropriate rotation matrix
so that A2 = M.A. A matrix which produces this particular set of rotations
is:
M =
_
0
1
2
1

3
2
_
4.2 Price Changes
Price movements are equivalent to the application of scaling operations which
can be modeled by the application of diagonal matrices. Thus a 50% fall in the
price of corn in our model would be represented by the application of the matrix
1 0
0
1

2
to the current commodity amplitude matrix. Scaling operations are not
length preserving.
The reason for this is that if there is a change in prices an agent holding
a vector of physical commodities b will nd that for most arbitrarily chosen
commodity vectors c, the quantity of c that they can exchange b for will have
changed. The length preserving rotations that we have assumed up to now have
amounted to assuming that there is no possibility of changing ones net worth
by commodity exchanges at a given set of relative prices. If the prices change
over time this is no longer the case, hence the introduction of non-conservative
operations.
Hilbert Space Models Commodity Exchanges 305
4.3 Modeling Debt
We specied in section 4 that the amplitude matrix must be complex valued.
This is required to model debt. Suppose that starting from holdings H agent
zero buys 2 of corn from agent one. Since agent zero only has 1 of gold to pay
for it, the transaction leaves the following holdings:
Agent gold corn
0 -1 2
1 2 2
The corresponding amplitude matrix is
A3 =
_
i

2

2
_
It it interesting that this too is the result of applying a unitary rotation
operator to the original amplitude vector since the length of row zero |A3
0
| =
i
2
+(

2)
2
= 1, likewise the lengths of all other rows and columns are preserved.
The linear operator required to create debts has itself to be complex valued, thus
if A3 = NA we have
N =
_
i
1

2
1

2
_
Note that the operators here are not Hermitian. This would appear to preclude
the interesting possibility of simulating commodity exchanges on a future quan-
tum computer[5], though there may be renormalization techniques that could be
applied to get over this problem.
5 Implications for Similarity Measures
Steedman [20] has proposed that a suitable criterion for assessing similarity of
values to market prices is the angle between market price and value vectors, with
small angles indicating closeness. If m, v are market price and value vectors
respectively, the angle between them is given by:
ArcCos(

m.

v )
where v denotes the normalized value vector given by v =
v
|v|
.
If the argument in section 2 is accepted, we should consider using angles
between price and value amplitude vectors instead. If we denote the normalized
vectors in amplitude space by m
a
and v
a
, then the amplitude space angles are
given by:
ArcCos(

m
a
.

v
a

)
where x

is the conjugate of x.
What will be the properties of this measure?
In general it will show smaller angles between vectors. For example suppose
we have 3 commodities iron, corn, cotton as follows:
306 P. Cockshott
amplitudes
value price value price
corn 1 1 1 1
iron 3 2

3

2
cotton 1 2 1

2
angle 30.2

13.4

The fact that smaller angles are shown would be or little signicance if the
relative sizes of angles in the two spaces was the same. But this need not be the
case. Consider the following example:
value price PP amplitudes
value price PP
corn 1 1 1 1 1 1
iron
1
2
-1 2
1

2
i

2
cotton 0.02 1 1

2
10
1 1
relative to price in 74

0 90

54

0 45

Here we are comparing three hypothetical vectors of values, prices and prices
of production (PP). If we treated commodity value space as a vector space, then
prices of production would be orthogonal to market prices, whereas in amplitude
space they are at 45

to market prices. In this example, when commodity space

is treated as a vector space, values appear closer to market prices than do prices
of production. When the assumption that commodity space is a vector space is
dropped, then prices of production are closer to market prices.
6 Conclusion
We have argued that commodity space can not be directly modeled by a vec-
tor space, because of the metric it observes, but that it can be treated as the
real valued representation or an underlying vector space. This complex vec-
tor space we have, following physics terminology, termed commodity amplitude
space. Observed holdings of commodities and money by agents are the squares of
corresponding commodity amplitudes. Commodity exchange relations, including
the formation of commercial debt can be modeled by unitary rotation matrices
operating on this amplitude space. The conceptual model presented borrows
extensively from quantum formalism.
It is thus at least arguable that the empirical relation between market prices
and labour values should be measured by the angles between their corresponding
vectors in commodity amplitude space. The latter space, unlike commodity space,
is a linear vector space within which angles of rotation have a clear meaning.
References
1. Cockshott, P., Cottrell, A.: Values Law, Values Metric, research report Univer-
sity of Strathclyde, Dept. of Computer Science (1994); Reprinted in: Freeman, A.,
Kliman, A., Wells, J. (eds.): The New Value Controversy and the Foundations of
Economics, Edward Elgar, Cheltenham (2004)
Hilbert Space Models Commodity Exchanges 307
2. Cockshott, W.P., Cottrell, A.: Labour-Time Versus Alternative Value Bases: A
Research Note. Cambridge Journal of Economics 27, 749754 (1997)
3. Cockshott, W.P., Cottrell, A.: A note on the organic composition of capital and
prot rates. Cambridge Journal of Economics 21, 545549 (2003)
4. Cockshott, W.P., Cottrell, A., Michaelson, G.: Testing Marx: Some New Results
from UK Data. Capital and Class 55, 103129 (Spring 1995)
5. DiVincenzo, D.: Quantum Computation. Science 270, 255261
6. Farjoun, E., Machover, M.: The laws of Chaos. Verso (1983)
7. Iverson, K.: A programming language. Wiley, New York (1966)
8. Iverson, K.: Notation as a tool of thought. In: ACM Turing award lectures, p. 1979.
ACM, New York (2007)
9. Kantorovich, L.V.: Mathematical Methods of Organizing and Planning Production.
Management Science 6(4), 366422 (1960)
10. Kliman, A.: The Law of Value and Laws of Statistics: Sectoral Values and Prices in
the US Economy, 1977-1997. Cambridge Journal of Economics 26, 299311 (2002)
11. Kurz, H.D., Salvadori, N.: Von Neumanns Growth Model and the Classical Tradi-
tion. Understanding classical Economics: Studies in Long-Period Theory (1998)
12. Ochoa, E.M.: Values, prices, and wageprot curves in the US economy. Cambridge
Journal of Economics 13(3), 413429 (1989)
13. Petrovic, P.: The deviation of production prices from labour values: some method-
ology and empirical evidence. Cambridge Journal of Economics 11(3), 197210
(1987)
14. Remak, R.: Kann die Volkswirtschaftslehre eine exakte Wissenschaft werden.
Jahrb ucher f ur Nationalokonomie und Statistik 131, 703735 (1929)
15. Ryabev, L.D., Smirnov, Y.N.: The Atomic Project, Science, and the Atomic In-
dustry. Atomic Energy 99(2), 519527 (2005)
16. Shaik, A.: The Empirical Strength of the Labour Theory of Value. In: Bellaore,
R. (ed.) Marxian Economics. A reapraisal Essays on Volume III of Capital. Prots,
Prices and Dynamics, pp. 225251. St. Martins Press, New York (1998)
17. Shaikh, A.: The transformation from Marx to Sraa. In: Freeman, A., Mandel, E.
(eds.) Ricardo, Marx, Sraa, pp. 4384. Verso, London (1984)
18. Sraa, P.: Production of commodities by means of commodities. Cambridge Uni-
versity Press, Cambridge (1960)
19. Valle Baeza, A.: Correspondence between labor values and prices: a new approach.
Review of Radical Political Economics 26(2), 5766 (1994)
20. Steedman, I., Tomkins, J.: On measuring the deviation of prices from values. Cam-
bridge Journal of Economics 22, 359369 (1998)
21. von Neumann, J.: A Model of General Economic Equilibrium. Review of Economic
Studies 13(33), 19 (1945)
22. von Neumann, J.: Mathematical Foundations of Quantum Mechanics, Engl. transl.
of the 1931 German edition by Beyer, R.T. (1955)