Randall Dougherty

Networks, Matroids, and Non-Shannon Information Inequalities

We define a class of networks, called matroidal networks, which includes as special cases all scalar-linearly solvable networks, and in particular solvable multicast networks. We then present a method for constructing matroidal networks from known matroids. We specifically construct networks that play an important role in proving results in the literature, such as the insufficiency of linear network coding and the unachievability of network coding capacity. We also construct a new network, from the Vamos matroid, which we call the Vamos network, and use it to prove that Shannon-type information inequalities are in general not sufficient for computing network coding capacities. To accomplish this, we obtain a capacity upper bound for the Vamos network using a non-Shannon-type information inequality discovered in 1998 by Zhang and Yeung, and then show that it is smaller than any such bound derived from Shannon-type information inequalities. This is the first application of a non-Shannon-type inequality to network coding. We also compute the exact routing capacity and linear coding capacity of the Vamos network. Finally, using a variation of the Vamos network, we prove that Shannon-type information inequalities are insufficient even for computing network coding capacities of multiple-unicast networks.

Unachievability of network coding capacity

The coding capacity of a network is the supremum of ratios k/n, for which there exists a fractional (k,n) coding solution, where k is the source message dimension and n is the maximum edge dimension. The coding capacity is referred to as routing capacity in the case when only routing is allowed. A network is said to achieve its capacity if there is some fractional (k,n) solution for which k/n equals the capacity. The routing capacity is known to be achievable for arbitrary networks. We give an example of a network whose coding capacity (which is 1) cannot be achieved by a network code. We do this by constructing two networks, one of which is solvable if and only if the alphabet size is odd, and the other of which is solvable if and only if the alphabet size is a power of 2. No linearity assumptions are made.

Nonreversibility and Equivalent Constructions of Multiple-Unicast Networks

We prove that for any finite directed acyclic network, there exists a corresponding multiple-unicast network, such that for every alphabet, each network is solvable if and only if the other is solvable, and, for every finite-field alphabet, each network is linearly solvable if and only if the other is linearly solvable. The proof is constructive and creates an extension of the original network by adding exactly s+5m(r-1) new nodes where, in the original network, m is the number of messages, r is the average number of receiver nodes demanding each source message, and s is the number of messages emitted by more than one source. The construction is then used to create a solvable multiple-unicast network which becomes unsolvable over every alphabet size if all of its edge directions are reversed and if the roles of source-receiver pairs are reversed

Linearity and solvability in multicast networks

It is known that for every solvable multicast network, there exists a large enough finite-field alphabet such that a scalar linear solution exists. We prove: i) every binary solvable multicast network with at most two messages has a binary scalar linear solution; ii) for more than two messages, not every binary solvable multicast network has a binary scalar linear solution; iii) a multicast network that has a solution for a given alphabet might not have a solution for all larger alphabets.

Network routing capacity

We define the routing capacity of a network to be the supremum of all possible fractional message throughputs achievable by routing. We prove that the routing capacity of every network is achievable and rational, we present an algorithm for its computation, and we prove that every non-negative rational number is the routing capacity of some network. We also determine the routing capacity for various example networks. Finally, we discuss the extension of routing capacity to fractional coding solutions and show that the coding capacity of a network is independent of the alphabet used

Network Coding and Matroid Theory

Networks derived from matroids have played a fundamental role in proving theoretical results about the limits of network coding. In this tutorial paper, we review many connections between matroids and network coding theory, with specific emphasis on network solvability, admissible network alphabet sizes, linear coding, and network capacity.

Critical points in an algebra of elementary embeddings

In this paper, we continue the study of a left-distributive algebra of elementary embeddings from the collection of sets of rank less thanto itself, as well as related finite left-distributive algebras (which can be defined without reference to large cardinals). In particular, we look at the critical points (least ordinals moved) of the elementary embeddings; simple statements about these ordinals can be reformulated as purely algebraic statements concerning the left distributive law. Previously, lower bounds on the number of critical points have been used to show that certain such algebraic statements, known to follow from large cardinals, require more than Primitive Recursive Arithmetic to prove. Here we present the first few steps of a program that, if it can be carried to completion, should give exact computations of the number of critical points, thereby showing that hypotheses only slightly beyond Primitive Recursive Arithmetic would suffice to prove the aforementioned algebraic statements.

Reducibility and nonreducibility between ℓ^{} equivalence relations

We show that, for 1 ≤ p < q < ∞, the relation of `p-equivalence between infinite sequences of real numbers is Borel reducible to the relation of `q-equivalence (i.e., the Borel cardinality of the quotient RN/`p is no larger than that of RN/`q), but not vice versa. The Borel reduction is constructed using variants of the triadic Koch snowflake curve; the nonreducibility in the other direction is proved by taking a putative Borel reduction, refining it to a reduction map that is not only continuous but ‘modular,’ and using this nicer map to derive a contradiction.

Insufficiency of linear coding in network information flow

It is known that every solvable multicast network has a scalar linear solution over a sufficiently large finite-field alphabet. It is also known that this result does not generalize to arbitrary networks. There are several examples in the literature of solvable networks with no scalar linear solution over any finite field. However, each example has a linear solution for some vector dimension greater than one. It has been conjectured that every solvable network has a linear solution over some finite-field alphabet and some vector dimension. We provide a counterexample to this conjecture. We also show that if a network has no linear solution over any finite field, then it has no linear solution over any finite commutative ring with identity. Our counterexample network has no linear solution even in the more general algebraic context of modules, which includes as special cases all finite rings and Abelian groups. Furthermore, we show that the network coding capacity of this network is strictly greater than the maximum linear coding capacity over any finite field (exactly 10% greater), so the network is not even asymptotically linearly solvable. It follows that, even for more general versions of linearity such as convolutional coding, filter-bank coding, or linear time sharing, the network has no linear solution.

Computations of linear rank inequalities on six variables

It is known that information inequalities on four random variables cannot be generated from a finite list. For the analogous case of linear rank variables, it is known that they can be generated from a finite list for up to five variables, but this is not known for six or more variables. Here we present partial results of computations on six-variable linear rank inequalities, showing that the number of sharp inequalities (those which cannot be generated from other inequalities) is more than one billion (counting variable-permuted forms). The problem is too large for standard polytope computation software; we describe the techniques used to generate and verify the current list of inequalities and a correspondingly large list of representable polymatroids. We also describe observed properties of the inequalities (some of which are now proven general results).