AA3 – Networks

Project

AA3-5

Tropical Mechanism Design

Project Heads

Michael Joswig, Max Klimm

Project Members

Sylvain Spitz (HU) 

Project Duration

01.01.2019 – 31.12.2021

Located at

HU Berlin

Description

We study the design of revenue-maximizing mechanisms for selling multiple items. Applying a duality framework, there is a one-to-one correspondence between optimal mechanisms and certain tropical polynomials and rational functions that we want to study via ideas from algebraic statistics.

TOPIC AND BACKGROUND OF THE PROJECT

The question of how to sell one item to multiple bidders in a revenue-maximizing auction is well understood since the work of Roger Myerson in the 1980s.  It is used in many applications and on online platforms such as eBay. But if we want to sell multiple items at once, the problem becomes substantially harder and not much is known about the optimal auction mechanism.

DETERMINISTIC AUCTION MECHANISMS

Typically, an auction mechanism asks the participants for their bids and then allocates the items to the bidders as well as a price they have to pay. This method is called a deterministic auction mechanism. In contrast, we could introduce a lottery for some (or all) items and, after collecting the bids, allocate the items to the bidders up to a certain probability. Observations suggest that in general a revenue-maximizing auction has to introduce such lotteries. All the more so it is interesting to find cases, in which a deterministic auction is optimal.

Giannakopoulos and Koutsoupias examined the case, where the valuations, that each bidder has for the items, are drawn according to an uniform distribution. They showed, that the optimal auction mechanism for this setting and up to 6 items is deterministic and they conjectured that this holds true for an arbitrary number of items. We will explore their conjecture as well as their approach by applying other methods, especially from tropical geometry. In fact, the utility function of a deterministic auction corresponds to a tropical polynomial. Moreover, we want to use the insights we get by this method, to better understand the general case.

Straight Jacket Auction and Generalized Permutahedra

The deterministic mechanism given by Giannakopoulos and Koutsoupias is called the Straight Jacket Auction (SJA). It consists of a certain price schedule (p₁,…,p_n), where p_k designates the price for any bundle of k items. Such a price schedule divides the n-cube in regions depending on the term which maximizes the utility function u(x) = max{ Σ_j∈J x_j – p_|J] | J ⊂ {1,…,n} }. The regions are marked on the right by U_J, where J is the bundle of items that attain the maximum utility. Their volumetric properties are essential for the computation of the prices and the proof of optimality of the mechanism.

The regions are in fact covector cells of the tropical polynomial u(x) intersected with the n-cube. They can also be classified as generalized permutahedra, which is useful because those polytopes come up in a variety of applications and are therefore well studied. In this way, we can link their volume to the number of bipartite graphs including a perfect matching.

Project Webpages

Selected Publications

Michael Joswig, The Cayley trick for tropical hyper surfaces with a view toward Ricardian economics, Homological and computational methods in commutative algebra, 107-128, 2017.

Paul Dütting, Felix Fischer, Max Klimm, Revenue Gaps for Static and Dynamic Posted Pricing of Homogenous Goods, arXiv:1607.07105, 2019.

Selected Pictures