Possibly relevant: In the first chapter of Where Mathematics Comes From: How The Embodied Mind Brings Mathematics Into Being, Lakoff & Núñez discuss a few examples of individuals who have suffered some kind of neurological damage that partially or completely disables their ability to perform numerical calculations. With respect to this question, the most interesting example is probably this one:
Not only is rote calculation localized separately from basic arithmetic abilities but algebraic abilities are localized separately from the capacity for basic arithmetic. Dehaene (1997) cites a patient with a Ph.D. in chemistry who has acalculia, the inability to do basic arithmetic. For example, he cannot solve $2\cdot 3$, $7-3$, $9\div 3$, or $5\cdot 4$. Yet he can do abstract algebraic calculations. He can simplify $(a \cdot b) / (b \cdot a)$ into $1$ and $a \cdot a \cdot a$ into $a^3$, and could recognize that $(d/c) + a$ is not generally equal to $(d + a) / (c + a)$. Dehaene concludes that algebraic calculation and arithmetic calculation are processed in different brain regions.
The citation to Dehaene (1997) refers to the book The Number Sense. The relevant passage from that book seems to be p. 199:
Up to now, this book has been concerned only with elementary arithmetic. But what about more advanced mathematical abilities, such as algebra? Should we postulate yet other neuronal networks dedicated to them? Recent discoveries by the Austrian neuropsychologist Margarete Hittmair-Delazer seem to suggest so. She has found that acalculic patients do not necessarily lose their knowledge of algebra. One of her patients, like Mrs. B, lost his memory of addition and multiplication tables following a left subcortical lesion. Yet he could still recalculate arithmetic facts by using sophisticated mathematical recipes that indicated an excellent conceptual mastery of arithmetic. For instance, he could still solve $7 \times 8$ as $7 \times 10 - 7 \times 2$. Another patient, who had a Ph.D. in chemistry, had become acalculic to the point of failing to solve $2\times 3$, $7-3$, $9\div 3$, or $5\times 4$. He could nevertheless still execute abstract formal calculations. Judiciously making use of the commutativity, associativity,and distributivity of arithmetic operations, he was able to simplify $\frac{a \times b}{b \times a}$ into $1$ or $a\times a\times a$ into $a^3$, and he recognized that the equation $\frac{d}{c} + a = \frac{d+a}{c+a}$ is generally false. Although this issue has been the matter of very little research to date, these two cases suggest, against all intuition, that the neuronal circuits that hold algebraic knowledge must be largely independent of the networks involved in mental calculation.
Regarding these passages, a couple of comments are in order:
- I am not sure if acalculia and dyscalculia are the same phenomenon, or what exactly the relationship is between them. Nor is it clear from this brief excerpt whether the patient (the one with the Ph.D in Chemistry) developed acalculia in adulthood, for example as the result of some injury, or whether it was a condition he had all along. (Hittmair-Delazer's first patient suffered from a "left subcortical lesion", but the Ph.D in Chemistry is "another patient", of whom no details are provided.) The distinction seems significant, as there is (probably?) a big difference between someone learning algebraic reasoning despite suffering from acalculia, on the one hand, and someone retaining algebraic reasoning that they learned prior to the onset of acalculia.
- Algebraic reasoning and manipulation is one thing, but proof-based advanced-level mathematics at the level of an undergraduate analysis course is (maybe?) something else.
The primary source for this is probably one of the following two papers:
HITTMAIR-DELAZER, M., SAILER, U., and BENKE, T. Impaired arithmetic facts but intact conceptual knowledge – a single case study of dyscalculia. Cortex, 31: 139-147, 1995.
HITTMAIR-DELAZER, M., SEMENZA, C., and DENES, G. Concepts and facts in calculation. Brain, 117: 715-728, 1994.
Both of these papers are cited in the references of yet another paper:
Dehaene, S., & Cohen, L. (1997). Cerebral pathways for calculation: Double dissociation between rote verbal and quantitative knowledge of arithmetic. Cortex, 33(2), 219-250.
I suspect any of these three papers might have additional references that could be helpful in understanding these phenomena.