Einstein T-norm (Einstein Product and Sum)

T-norms are the mathematical counterpart of the connective "AND" in fuzzy and intuitionistic fuzzy set theory. They reduce two membership degrees, or the two components of an intuitionistic fuzzy number, to a single degree. The Einstein t-norm does this with a fixed rule that carries no adjustable parameter. It is not a decision method in its own right; within host methods such as intuitionistic fuzzy TOPSIS or intuitionistic fuzzy weighted averaging, it runs the step that combines the assessments of several experts or several criteria. Its output is not a ranking but a single number, or a single intuitionistic fuzzy number (a new membership–non-membership pair). Xu and Yager (2006) proposed this rule for adding and multiplying intuitionistic fuzzy numbers; it is preferred in studies wanting a slightly more cautious result than algebraic aggregation.

A t-norm's task is to answer the question: if two conditions are each satisfied to some degree, to what degree are both satisfied together? The best-known answer is the product; this is called the algebraic t-norm. The Einstein t-norm answers the same question a little more guardedly, and its result is always less than or equal to the algebraic product. It takes its name from the velocity-addition rule in relativity: there too, when combining two quantities, the result comes out slightly more "compressed" than ordinary addition and can never exceed an upper limit. In a decision context, this is the mathematics of a preference not to be overly optimistic when combining two independent pieces of evidence or two expert opinions. Its philosophical consequence is this: Einstein aggregation is less compensatory than algebraic aggregation. If one of two assessments is weak, the Einstein combination forgives that weakness less than the algebraic combination does.

The method proceeds through three steps.

First step, the Einstein product. When two values (say, the confidence degrees of two independent pieces of evidence) are combined, the calculation starts from the ordinary product but is adjusted so the result comes out slightly smaller than the ordinary product. For example, the ordinary product of 0.6 and 0.7 is 0.42; the Einstein product of the same two values is 0.375. The Einstein product is always less than or equal to the ordinary product.

Second step, the Einstein sum. This is, in a sense, the mirror image of the product: the two values are combined with an "either/or" logic, that is, a logic closer to the connective OR. The ordinary (probabilistic) sum of 0.6 and 0.7 is 0.88, while their Einstein sum is 0.915. The Einstein sum is always greater than or equal to the ordinary sum, but it eases off more gently as it approaches the upper limit neither can exceed, namely 1.

Third step, combining two intuitionistic fuzzy numbers. An intuitionistic fuzzy number consists of two degrees: membership, that is, how positive it is, and non-membership, that is, how negative it is. When combining two assessments (from two experts, or two criteria), the membership degrees are combined with the Einstein sum and the non-membership degrees with the Einstein product. Positive evidence thus combines under a more generous rule and negative evidence under a stricter one, and the result remains a valid intuitionistic fuzzy number.

The formulas behind each step, the intermediate tables and the citation formats are given on the DecisionMind method page; this card carries no formulas.

The Einstein t-norm's output is not, on its own, a decision result; it is an intermediate step within a host method. The output is either a single number resulting from combining two numbers, or a new membership–non-membership pair resulting from combining two intuitionistic fuzzy numbers. The score derived from this pair, that is, membership minus non-membership, shows the combined evidence's overall tendency. A score close to 1 means a strong positive tendency, and one close to -1 means a strong negative tendency. This score can be compared only with other scores computed under the same choice of t-norm. A score from algebraic aggregation and a score from Einstein aggregation may appear on the same scale, but they are products of different combination rules and cannot be set directly side by side.

For this reason:

"Einstein aggregation gave the true average"

should be written instead as:

"These assessments were reduced, under the Einstein t-norm's cautious combination rule, to this single intuitionistic fuzzy number; had the same data been combined under the algebraic rule, the score would have come out differently"

The Einstein t-norm works with intuitionistic fuzzy data: every assessment consists of a membership degree and a non-membership degree, and the two must not sum to more than 1. The same rule is also used for closely related data types such as Pythagorean fuzzy and q-rung orthopair fuzzy. DecisionMind holds no separate extension of this building block; the Einstein t-norm is identical to the special case of the Hamacher t-norm at γ=2. You need at least two intuitionistic fuzzy values: two expert opinions, two criterion assessments, or a single intuitionistic fuzzy number's own two components. The Einstein t-norm neither requires nor produces weights; weighting is handled separately within whichever host method it runs inside.

If your data is intuitionistic or Pythagorean fuzzy and you want to combine several pieces of evidence or expert opinions with a rule slightly more cautious than the algebraic product, the Einstein t-norm is a suitable choice. The cases where it should not be used are as follows: if your data is crisp, a t-norm is unnecessary and a direct weighted average suffices. If you want to tune the degree of caution to your data set, a parameterised family is needed rather than Einstein, which is a fixed rule. Where the choice of t-norm cannot be justified in the report, this choice also undermines transparency.

Intuitionistic fuzzy data, combination slightly more cautious than the algebraic product → Einstein t-norm

Same need, but the degree of caution should be tunable to the data → Hamacher t-norm (parameter γ)

Same need, a sharper cut-off wanted at the extremes → Schweizer-Sklar t-norm (parameter p)

Data is crisp, no combination rule is needed → a direct weighted average

The Einstein t-norm's most important advantage is its simplicity. It carries no parameter, which removes the "which parameter should be chosen" debate entirely. Because it is the γ=2 special case of the Hamacher family, it is well known in the literature both on its own and as a recognised member of a broader family. It curbs overly optimistic results compared with algebraic aggregation and is computed by a closed-form formula, requiring no iteration or optimisation.

Its limitations stem from this same simplicity. Because it carries no parameter, it cannot be tuned; its degree of caution is fixed and cannot be adjusted to a data set's characteristics, whereas the Hamacher and Schweizer-Sklar families can do this through a parameter (Klement, Mesiar and Pap, 2000). Second, on mid-sized values (the 0.4-0.6 range, say) it produces results very close to algebraic aggregation; in that case it becomes harder to offer a strong justification for "why Einstein was chosen." Third, there is no consensus in the literature on which t-norm is the "correct" rule for combination; the choice is mostly a matter of convention or of the software's default (Xia, Xu and Zhu, 2012).

The most common mistake is confusing the Einstein product with the ordinary product; the Einstein product is always less than or equal to the ordinary product, and can never be larger. A second mistake is confusing which operation applies to membership and which to non-membership. Membership combines with the Einstein sum and non-membership with the Einstein product; reversing this can produce an invalid intuitionistic fuzzy number whose sum exceeds 1. A third mistake is treating the Einstein t-norm as an independent method separate from Hamacher; Einstein is a single point in the Hamacher family at γ=2, and the relationship between them must be stated in the report.

The governing principle is this:

The Einstein t-norm does not determine an aggregation's result but the rule by which that result was produced; the report must name that rule explicitly.