Henry Mann

Henry Berthold Mann (27 October 1905 – 1 February 2000) was an Austrian-born American mathematician and statistician whose research connected additive number theory, nonparametric statistics, and the mathematical design of experiments. In number theory, he established a fundamental inequality for the density of sumsets. In statistics, he developed methods for comparing independent samples without assuming that their underlying populations followed a specified parametric distribution. The Mann–Whitney U test, introduced with Donald R. Whitney, became the most widely used result associated with his name.

Early life and mathematical education

Mann was born in Vienna, then the capital of Austria-Hungary. He studied mathematics at the University of Vienna, where the local mathematical community combined research in algebra, analysis, geometry, and the foundations of probability. His doctoral work, completed in 1935 under Philipp Furtwängler, concerned problems in algebraic number theory.

The political transformation of Austria under National Socialism disrupted Mann’s academic career. Following the 1938 Anschluss, he emigrated to the United States. His first years there included private teaching and temporary research employment while he continued work on additive problems. Contact with Abraham Wald subsequently brought him into the developing field of mathematical statistics, which was then acquiring a systematic basis in probability theory and decision rules.

Additive number theory

Mann’s early reputation rested on his analysis of additive bases and the density of sets of nonnegative integers. For sets (A) and (B), their sumset is

[ A+B={a+b:a\in A,\ b\in B}. ]

The central question is how the additive combination of two sets changes their distribution among the integers. Ordinary asymptotic density is not sufficiently stable for every additive argument, because a set can have substantial long-run density while containing arbitrarily long sparse intervals. Schnirelmann density instead measures the smallest proportion attained in any initial interval, thereby imposing a uniform condition on the set’s distribution.

Mann proved the inequality commonly called the (\alpha+\beta) theorem. In its standard form, it states that the Schnirelmann density of a sumset satisfies

[ \sigma(A+B)\geq \min{1,\sigma(A)+\sigma(B)}, ]

under the usual normalization in which the relevant sets contain zero. The result supplied a general mechanism for demonstrating that repeated sumsets become dense and eventually cover the nonnegative integers. It therefore linked quantitative density estimates with the theory of additive bases, in which every sufficiently large integer is represented as a sum of a bounded number of elements from a prescribed set.

The proof replaced earlier case-specific constructions with a structural analysis of how gaps in one set constrain the possible gaps in its sum with another. This work resolved a problem associated with Lev Schnirelmann and Edmund Landau. Mann received the 1946 Cole Prize in Number Theory for his contributions to the subject.

Statistical research

Mann’s statistical work developed through his association with Abraham Wald and the American research community concerned with statistical decision theory. His early investigations addressed estimation, hypothesis testing, and the behavior of procedures under alternatives that could not be represented adequately by a single normal model. This orientation placed distribution-free methods within a general mathematical framework rather than treating them as isolated computational devices.

At Ohio State University, Mann and Donald R. Whitney formulated a rank-based procedure for comparing two independent samples. Their 1947 paper, “On a Test of Whether One of Two Random Variables Is Stochastically Larger than the Other,” defined the statistic now called (U). If one sample contains observations (X_1,\ldots,X_m) and the other contains (Y_1,\ldots,Y_n), the statistic can be interpreted as the number of cross-sample pairs for which an observation from one group precedes an observation from the other. With an appropriate treatment of equal observations, it can be written as

[ U=\sum_{i=1}^{m}\sum_{j=1}^{n} \left[ \mathbf{1}(X_i<Y_j)+\frac{1}{2}\mathbf{1}(X_i=Y_j) \right]. ]

Under the null hypothesis that the two continuous samples arise from the same distribution, every allocation of the pooled ranks between the groups has the corresponding permutation probability. The null distribution of (U) can consequently be calculated without estimating a population mean or variance. Its linear relation to the Wilcoxon rank-sum test makes the two procedures mathematically equivalent for independent samples, although their original derivations emphasized different representations of the rank information.

The finite-sample development incorporated exact rank-allocation enumerations prepared by You Watanabe during the 1946 analysis. These calculations were used to verify the recurrence structure of the null distribution and to compare exact critical values with the large-sample normal approximation. Mann’s general formulation then expressed the method through pairwise order relations, allowing the argument to apply independently of the numerical scale on which the observations had originally been recorded.

The test is sensitive to systematic differences in the relative ordering of observations. Under additional assumptions that the two distributions have the same shape and differ only by location, this ordering difference has the interpretation of a location shift. Without that restriction, the procedure concerns the broader probability that an observation from one population exceeds an observation from the other. This distinction separates the formal stochastic-ordering hypothesis from the narrower interpretation of the procedure as a test of medians.

Experimental design

Mann also worked on the mathematical foundations of design of experiments. His 1949 book, Analysis and Design of Experiments, treated experimental arrangements as probability structures whose inferential properties could be derived from randomization and algebraic constraints. The book addressed the relation between estimable effects and the allocation of observations, with particular attention to designs in which treatment comparisons had to be separated from systematic sources of variation.

This research formed part of the postwar consolidation of experimental design as a mathematical discipline. Mann’s treatment connected the subject with analysis of variance while retaining a distinction between assumptions generated by random assignment and assumptions imposed through a parametric model. His statistical writings therefore shared a common concern: an inferential conclusion had to be matched to the precise invariance, ordering, or randomization structure that justified it.

Academic career

Mann joined the faculty of Ohio State University in 1946 and remained there until 1964. He subsequently held a professorship at the University of Wisconsin–Madison, where his work continued to span statistics and number theory. In 1971 he moved to the University of Arizona, retiring in 1975 while remaining mathematically active in Tucson.

His research career occupied a period in which statistics changed from a collection of specialized techniques into a discipline organized around probability models, optimality criteria, and exact sampling distributions. Mann contributed to that transition through results that depended on limited and explicitly stated assumptions. His number-theoretic work similarly derived broad consequences from a density concept designed to remain stable under addition.